I’ve done performance tuning long enough to develop some rules of thumb:
Just about everyone will detect a 3x performance difference if told about it. (5x even if they’re not told.)
Only gamers and some IT pros will notice a 50% difference.
Nobody will notice 20% or less.
Anything under 10% is hard to even measure.
One career trick is that if you have three changes that provide a 20% boost each, release them at the same time instead of trickling them out one at a time. Sure, this is more risky for the company but is great at review time because otherwise your contributions would go unnoticed. Unethical? Maybe, but this is the incentive structure. Hate the game, not the player.
There is also an adaptation factor: Good performance is only noticeable in contrast to poor performance. You get used to it very fast and stop noticing, unless you go back to the old system which suddenly feels "broken", even though you may have considered it perfectly acceptable before.
For example, if the first visible paragraphs takes the same time to show up, but the rest of the page is slower (but still not slow enough to be visible even if the user starts to scroll right away), then there is objectively 0 improvement on user interaction
This obviously revolves around what one perceives. Data speeds on mobile are objectively faster on their newer devices [1]. But we normalise those speeds without needing additional cruft; waiting for a page to load just ceases to be a thing one notices except when it doesn't work.
[1] https://www.opensignal.com/2023/09/07/users-should-upgrade-t...