Protocol Gains, Payload Losses
HTTP/2 was published as a standard in 2015 to modernize an aging protocol, and HTTP/1 had turned performance work into an arcane discipline of workarounds for its limitations. HTTP/2 adoption still isn't absolute and prioritization issues remain unresolved, but the web is clearly better off for it. Yet the years since its rollout also brought a 102% median increase in bytes transferred over mobile. At the 90th percentile of the same dataset — where the long tail of performance lives — the increase is 239%. Over roughly the same period, average mobile download speeds in the U.S. rose 73%; Brazil and India saw 75% and 28%. Faster networks and better protocols have not been matched by restraint in what we ship.
Page weight alone doesn't tell the whole story of the user experience, but it is at minimum loosely tied to a collective problem. CrUX, the Chrome User Experience Export, feeds HTTPArchive's reporting, and while that data can be read several ways, one fact holds: most CrUX metrics over the last couple of years show little if any improvement, despite gains in browsers, the HTTP protocol and the network itself. The best that can be said of these improvements is that they have stemmed the tide of our excesses, not reduced them. We keep building in a way that sustains a never-ending Jevons paradox.
Why Performance Stalls
Several forces stand in the way of a faster web for everyone:
- The relentless desire to monetize every square inch of the web, along with the army of third-party vendors that such efforts fund.
- Workplace cultures that favor unrestrained feature-driven development — a practice that adds to, but rarely subtracts from, what gets sent to users.
- Developer conveniences that ease the developer's job while placing increasing cost on the client.
Counter-intuitively, owners of mature codebases exhibiting these traits keep walking the same unsustainable path to profitability rather than acknowledge the repeatedly established fact that performance-first development does as much — or more — for the bottom line and the user experience.
What we call performance remediation is largely a set of engineering techniques responding to the side effects of business, product management and engineering practice. We are good at applying tourniquets and poor at sewing up deep wounds.
Performance Is a People Problem
Technical solutions are comparatively inarguable: content compression works, minification works, tree shaking works, code splitting works. The intersection of performance and people is messier. Unlike an obviously beneficial technical fix such as HTTP/2, how do you qualify what a successful performance culture looks like, or how to get there? The following combination of cultural and engineering tenets is a reasonable template:
- Leadership support. An organization can't prioritize performance without its leaders behind it. Without that, building a culture where performance is the product's primary feature is extremely difficult.
- Telemetry. Even with leadership backing, performance can't be prioritized effectively if it isn't measured. Without measurement you can't explain how product development affects performance, and no one cares until there's an apparent crisis.
- Organizational understanding. Leadership support plus telemetry still isn't enough unless the whole organization understands web performance. This is where training, documentation, best practices and standards come in — an area many organizations have already invested in. The harder work is establishing feedback loops to assess how well that knowledge is understood and applied.
- Accountability. Once the other pieces are in place, accountability becomes possible — not as reprisals when telemetry shows performance has suffered, but as guard rails in the deployment process that flag when thresholds are crossed.
Even if all of these things come together in your workplace, good outcomes aren't guaranteed.
Absent regulation forcing us to address poorly performing websites — the way the ADA keeps accessibility in focus — maintaining performance as a priority will take continuing evangelism and pressure. Work on the user experience in evolving codebases is never finished.
No Resting on HTTP/3 and 5G
As HTTP/3 and 5G emerge, we must not assume they will heal our ills once and for all. If we do, we'll be having this discussion again when their successors loom. Innovation alone can't keep the web fast, because making it fast — and keeping it that way — is hard work that only happens when we do it together.



