At the time I was most bothered by the slow load times of web pages and blaming this phenomenon:
https://www.sjsu.edu/faculty/watkins/samplemax4.htm
particularly that if you take the max of N random variables, the expectation value you get gets worse as N increases -- that is, the page isn't done loading until the slowest http request completes.
So I saw the "knowing when the page is done" problem as being particularly core, and it would be if the goal was to "win the race" against a conventional web browser.
If you were (say) preloading all the links submitted to hacker news you might be able to tolerate the system taking 5 minutes to process an incoming page. (See archive.is)
Today I've noticed that sites like Wired are giving up on complaining about my anti-track and ad-blocker and they just load the page partially which would drive me crazy if I was serious about debugging.