Journal · 12 June 2026

HTTP/3 in practice: what changed, what did not

We enabled HTTP/3 on a handful of small sites last spring and collected a year of timing data. The honest summary: it helps in specific conditions and does almost nothing for the problems most sites actually have.

Where it helped

Visitors on mobile networks with a changing IP — stepping off Wi-Fi onto LTE mid-page-load — saw fewer interrupted requests. Connection migration is real and measurable, though it affects a small slice of sessions.

Where it did not

Median first contentful paint barely moved. The dominant factors remained the same: render-blocking CSS, oversized images, and third-party scripts loaded before the main content. On one project, removing a single tag manager saved more time than the protocol change by an order of magnitude.

What we tell clients

Turn it on if the host supports it without extra cost; do not make it a project goal. Measure transfer size and blocking resources first. A protocol upgrade is not a substitute for deleting 300 KB of JavaScript nobody uses.

← Back to the journal