Why the web's most beautiful sites have the worst performance scores
The 2024 Site of the Year scores 59/100 on Google's performance test with 10.4 seconds of blocking. Five award-winning sites audited for this article tell a different story than "beautiful means slow": what the scores actually measure, what they ignore and how to make the call for your own site.
59 out of 100. That is the mobile performance score of the site named Site of the Year 2024 by Awwwards (opens in a new tab), the most watched prize in web design. The test was run on July 31, 2026 with Lighthouse, Google's audit tool: 10.4 seconds of browser blocking, 16.9 MB transferred. Seen from a dashboard, a disaster.
Yet the same report contains two excellent results. The largest visible element renders in 1.8 seconds, under the 2.5-second threshold that Google recommends (opens in a new tab). The layout does not move a single pixel during load: a cumulative shift of zero, the perfect score. A site packed with real-time 3D therefore posts flawless results on two of the three official metrics.
This gap inside a single report is the real subject. It reveals what performance scores actually measure, what they ignore and why the question "do you have to choose between a remarkable site and a fast site" is badly framed. To answer it with facts rather than impressions, five of the most awarded sites in the world were audited on the same day, with the same tool and the same settings.
Five award-winning sites, one audit
The panel: two Awwwards Sites of the Year, Igloo Inc (opens in a new tab) (2024) and Lusion (opens in a new tab) (2023), plus three reference studios of the immersive web, Active Theory (opens in a new tab), Locomotive (opens in a new tab) and Obys (opens in a new tab). The audit: Lighthouse in mobile emulation, meaning a slow 4G network and a CPU throttled to a quarter of its speed, the tool's standard methodology (opens in a new tab). Anyone can reproduce these measurements in five minutes from PageSpeed Insights (Google's online tester) or Chrome DevTools.
Before reading the table, the four columns in plain words:
- Score /100: the overall grade Lighthouse displays in green, orange or red. An aggregate of several metrics, each with its own weight in the grade.
- First visual (LCP, Largest Contentful Paint): the time before the largest element of the page is displayed, an image, a heading or a video. It is the moment the visitor feels the page has arrived. Recommended threshold: 2.5 seconds.
- Blocking (TBT, Total Blocking Time): the browser does almost everything in a single queue, the main thread: running code, drawing, reacting to clicks. TBT adds up the moments during load when that queue is too busy to respond. A page in full blocking is visible but deaf: you touch the screen, nothing reacts.
- Stability (CLS, Cumulative Layout Shift): the extent of layout jumps during load, the button that slips away under your finger because an image just pushed everything down. Zero means nothing moves. Recommended threshold: 0.1.
| Site | Score /100 | First visual (LCP) | Blocking (TBT) | Stability (CLS) | Weight |
|---|---|---|---|---|---|
| Igloo Inc | 59 | 1.8 s | 10,400 ms | 0 | 16.9 MB |
| Lusion | 54 | 4.7 s | 860 ms | 0 | 10.1 MB |
| Active Theory | 39 | 5.4 s | 11,550 ms | 0 | 16.6 MB |
| Locomotive | 10 | 8.3 s | 2,280 ms | 1.518 | 9.3 MB |
| Obys | 63 | 6.1 s | 50 ms | 0.279 | 0.96 MB |
Measured on July 31, 2026, Lighthouse 13.4 in mobile emulation, one simulated visit per site. The order of magnitude is reliable, the decimal is not.
Three observations come out of this table and none of them matches the "beautiful means slow" reflex.
- Weight does not predict experience. The heaviest site in the panel renders its first visual in 1.8 seconds, faster than the lightest one. And that lightest site proves something else: an award-winning art direction can fit in 0.96 MB. What weighs is the 3D worlds, not the design.
- Stability is a design decision, not an optimization. Three sites saturated with animation post a layout shift of zero: their movement relies on transforms, displacements computed by the graphics card that never push the surrounding content around. Conversely, a CLS of 1.518 signals a page that jumps during load, fifteen times the recommended threshold.
- The real conflict is concentrated on a single dimension: interactivity. The two blocking times above 10 seconds come from the two sites running a full 3D world in the browser. It is the only place where visual ambition and the metric truly collide.
The score everyone screenshots is not the one that counts
The number out of 100 is a lab result: one simulated visit, on a simulated mid-range device, over a simulated degraded network. The Core Web Vitals come from somewhere else. Google's three official metrics (rendering with LCP, responsiveness with INP, stability with CLS) are drawn from the Chrome UX Report (opens in a new tab), which aggregates real visits from Chrome users on real devices. The official verdict is decided at the 75th percentile of those real visits, in other words: the threshold must hold for at least three visits out of four. A site viewed mostly on desktop, by an audience that came specifically for the experience, can therefore fail the mobile lab test and pass the thresholds in the field. The lab explains, the field decides.
The score's composition amplifies the misunderstanding. In the current version of Lighthouse, browser blocking (TBT) weighs 30% of the grade, more than any other metric. A 3D site therefore loses nearly a third of its score on the exact dimension where its technology costs the most, even when its rendering and stability are exemplary. Igloo Inc's 59/100 reads like this: two perfect metrics, one massive and concentrated penalty.
Another subtlety, a decisive one: the official interactivity metric, INP (opens in a new tab), cannot be measured in a lab. It observes the latency of real interactions (clicks, taps, keystrokes) across the whole visit, with a 200-millisecond threshold. The lab's Total Blocking Time (opens in a new tab) is only a stand-in: it measures blocking during load, not a failed interaction. The 10,400 ms in the table signal a risk, not a conviction. If the user touches the screen during that blocking, nothing responds. If they are watching an opening animation at that moment, they never meet the problem.
Award-winning studios build precisely on this. The intro screen, often mocked as an aesthetic whim, occupies the exact seconds when the 3D engine loads its textures and compiles its shaders, the programs that draw the scene on the graphics card. The blocking exists, the visitor never crosses it. That is design, not cheating: incompressible time is turned into staging instead of being endured as a blank page.
What Google does with it, what your business does with it
On the search side, the official position is more measured than the panic it inspires. John Mueller, Google Search's spokesperson, said it again in 2024 (opens in a new tab): Core Web Vitals "are not giant factors in ranking" and content relevance weighs far more. A remarkable site does not vanish from results because its audit is orange. The detailed SEO mechanics are covered in the article on Core Web Vitals in 2026 (opens in a new tab).
The real stake is elsewhere: conversion. The Milliseconds Make Millions (opens in a new tab) study, run by Deloitte and the agency 55, tracked 37 brand sites and over 30 million sessions. Its result fits in a tenth of a second: a 0.1 s improvement in mobile speed lifts retail conversions by 8.4% and average order value by 9.2%. On a site that sells, performance is not a technical grade. It is a line on the income statement.
So why do award-winning studios accept average scores? Because their site's contract is not the one a business site signs. A studio portfolio is a demonstration: visited deliberately, often on a good machine, by an audience that came for the experience and is willing to wait through a curtain-raiser. A business site receives hurried visitors, on mobile, in the middle of comparing three providers. The former can stage ten seconds of waiting. The latter pays for every second in lost prospects.
Three thresholds, three different negotiations
The useful reading grid comes straight out of the table: the three Core Web Vitals do not punish the same choices and two out of three are fully compatible with a spectacular site.
- LCP, 2.5 s: holdable even at 17 MB. The threshold punishes a lazy first visual, not the richness of what follows. Proof by the Site of the Year 2024: 1.8 seconds. The method is constant: display a light, immediate composition first, load the 3D engine behind it.
- CLS, 0.1: non-negotiable, ever. No creative ambition justifies a page that jumps. The most animated sites in the panel sit at zero: transform-based animation does not displace content and therefore shifts nothing. A high CLS does not betray boldness, it betrays unreserved space or a badly loaded font.
- INP, 200 ms: the only real negotiating ground. This is where 3D, heavy JavaScript and rich interactions actually cost. So this is where the craft happens: animations handed to the compositor (the part of the browser that moves pixels without going through the main thread), code split, loads deferred, the queue freed between interactions.
The precise techniques are covered in dedicated articles: rendering and main thread work (opens in a new tab), animations handed to CSS instead of JavaScript (opens in a new tab) and building in progressive layers (opens in a new tab). The essential fits in one sentence: the spectacular loads after the usable, never instead of it.
Deciding based on your site's contract
| Site type | Performance requirement | Creative margin |
|---|---|---|
| Showcase or business site | Green Core Web Vitals in the field, no exception: the audience is in a hurry and comparing | Full on rendering and stability. Rich interactivity concentrates on the moments that guide or reassure |
| Immersive brand site | LCP and CLS held (the proof exists), INP monitored on action paths: menu, contact, form | Maximal, including an assumed curtain-raiser |
| Portfolio, event site | The field decides: if the audience comes for the experience, the lab score is a data point, not a verdict | Total |
One last reflex, the most important: measure on real visitors rather than on the score. Field data is freely available in Google's Search Console and the Chrome UX Report as soon as traffic allows. As for knowing whether an animation earns its place in the experience, the method exists: it is the subject of the article on measuring animations (opens in a new tab).
The paradox is not one
The web's most beautiful sites do not have bad scores because beauty would be expensive. They concentrate their spending on a single one of the three official dimensions, the one their audience agrees to pay. On the other two, the best are exemplary, with weight budgets nobody would dare recommend.
A lab score measures one simulated visit on a mid-range device. It measures neither memorability, nor singularity, nor the reason a visitor mentions a site a week later. Those things cannot be read in a report: they are built, by knowing exactly where every millisecond goes. That precision is what separates a singular site from a merely heavy one.