Quote from the article: "Rendering time dropped from about 4200ms in native AngularJs to 120ms using ReactJs." So just above 4 seconds. And yes, if you have an app where only one in 10000 pages need that sort of performance it might not be necessary. But it's a one time (per user) cost of 32KB download to improve the performance of the "challenging" parts of your app. Which can make a big difference in the perception of users. Especially since we didn't say anything about the distribution - if the 95% percentile is still 2+ seconds, the %.01 don't really matter.

I've seen serious performance problems with Angular for even reasonably sized lists (granted, with fairly complex display logic) that React handled without making the browser hang and then scrolling lag noticeably. So I'd personally rather drop the 36KB of Angular before dropping the 32KB of React. ;) But if someone is just displaying a handful of elements with simple logic and is happy with Angular's performance, then there's no need to frankenstein React into it.

heh yeah i don't really have performance benchmarks but i went React over Angular -- I've found the style to be a little odd (passing bound handler functions through descendants to update a state param in the grandparent class? surely they can prescribe something better) but the performance is perfect & the UI I'm working on is rather complex (a few hundred nested arrays of js objects that are themselves about 8 data points, with a lot of state-dependent display rules).

I feel like maybe Angular would be more convenient to program as an app, but I still have server-side logic -- my priority with a js lib is just to have a highly interactive UI even if it's just a fairly dumb state machine. To be honest, I feel like "dumb" has helped. I put just enough into state, just enough into events without being super model-oriented. Incidentally, I hear that is why Angular's perforance is poor (executes a bunch of js on model behavior even if it only results in 1 render -- not sure if this is true or not but I saw some interesting graphics of browser profiling that showed a ton of methods firing just for 1 small model update).