Measured, not argued
How fast is your website platform, really?
Not a lab test on one page. This is what real Chrome users experienced across millions of real websites — grouped so that a product is only ever compared with a product.
The more of the stack somebody else runs, the more often a site passes. That is the pattern in those three numbers — and it is the honest argument for a hosted builder over a custom build, if all you need is five pages.
Self-hosted software
A finished product with an admin screen, running on hosting you chose. The software is somebody else's; the caching, the images and the server are yours to get right.
- OpenCart73.7%20,404 sites
- Drupal64.6%93,208 sites
- PrestaShop59.8%49,523 sites
- Joomla57.1%87,332 sites
- WordPress48.7%2,752,367 sites
- Magento43.9%32,992 sites
OpenCart. 1st of 6 in this tier, 73.7% passing across 20,404 measured sites.
- 84%LCP · appearshow long until the main content shows up
- 90%CLS · holds stillhow much the layout jumps while loading
- 94%INP · respondshow long a tap or keypress takes to register
No single metric stands out for OpenCart: it sits near the middle of the field on all three.
Share of origins passing all Core Web Vitals on mobile, August 2026, across 6,567,132 measured websites. Chrome UX Report field data — real visits by real people, not a lab test — published by the HTTP Archive under CC BY 4.0. Which 23 technologies are listed is my choice — the dataset covers many more. The numbers are not: each one is here only because the dataset measured at least 20,000 of its origins, below which the figures swing too much month to month to rank.
And the limit of it. This is not one website built 23 ways. Hosted platforms mostly carry small brochure sites their vendor tunes end to end; frameworks mostly carry shops, applications and publications, loaded with tag managers and analytics — which is exactly what ruins INP. So a tier median says what tends to get built this way, not how fast the technology can go. Astro at 71% and React at 45% are both frameworks; the difference is how much JavaScript reaches the browser. This site is built with Next.js, which the dataset counts as React — the slower one.