The problem however is that JS optimization is very sensitive, depending on the browser vendor, version, and execution context, optimization advice can change dramatically.

Back when I worked on a perf sensitive JS codebase, we had a perf test suite, and depending on the details, often times locals were faster, but sometimes objects, and on a different browser it was arrays. And if you accident hit an ask.js opt it could be typed arrays, but mostly typed arrays were significantly slower. And when a new version came out this could all invert.