Truth be told, I will still apply some of these optimizations to code that I know will run very frequently. For instance, in the virtual list computation logic I wrote for Pinafore, I was careful to use only for-loops and not Array.map(), Array.filter(), etc.

Ideally, you'd only worry about this stuff if perf testing actually revealed that it was a problem. But on the web, you'd have to measure on multiple browsers/devices to be sure, which isn't always feasible. Sometimes hunches are okay.

This weekend I played around with Cypress and TestCafe for e2e testing, but ultimately went with TestCafe because:

1. it supports more browsers than just Chrome
2. it ran my tests faster than Cypress (even without concurrency)
3. the concept of "roles" is really useful for mocking login behavior and ensuring you only need to log in the test user once: https://devexpress.github.io/testcafe/documentation/test-api/authentication/user-roles.html

I will admit Cypress has the better interactive test environment, though.