HTTP/1.1 opened a limited number of parallel connections per domain, which is why concatenating every CSS file into one and spriting every icon became standard practice - fewer requests meant less time queued waiting for a free connection.

Multiplexing removes the queue

HTTP/2 multiplexes many requests over a single connection, so the old penalty for "too many small files" mostly disappears. Bundling everything into one giant CSS file can now actively hurt - a small change forces the whole bundle to be re-downloaded, where several smaller, well-cached files wouldn't need to be.

# Serving over HTTP/2 (nginx)
listen 443 ssl http2;

Server push, used carefully

HTTP/2 also lets a server push resources the browser hasn't asked for yet. It's powerful and easy to misuse - pushing something the browser already has cached wastes bandwidth - so we've held off using it broadly until the tooling for cache-aware push matures.

Multiplexing quietly undid a decade of workarounds

HTTP/1.1's one-request-per-connection limit is why sprite sheets, file concatenation, and domain sharding became standard practice - they were all ways to fight a browser connection limit that no longer applies under HTTP/2's multiplexed streams over a single connection.

/* HTTP/1.1 workaround: bundle everything into one file */
/* HTTP/2: many small, cacheable files load in parallel just fine */

Optimization advice that quietly became outdated

Concatenating every CSS and JS file into one giant bundle actually hurts caching under HTTP/2 - a single character change invalidates the whole bundle, whereas smaller, well-scoped files let browsers cache the parts that didn't change. A lot of "best practice" performance advice from the HTTP/1.1 era needs re-examining under HTTP/2 and HTTP/3, since the underlying constraints those practices were solving no longer exist.