At what point do you stop overthinking random code optimizations and move on to the next task?

now

I usually go with, first make it work then make it go fast if necessary, preferrably you should get a benchmark from the first solution so you can actually prove the improvement

If you're not measuring performance, you're not optimising. If you don't have a target performance, there's no need to optimise, so just make it work. Otherwise, if you're actually working on a performance critical piece of work, you need to be profiling, then it's only worth optimizing the top few functions in terms of runtime cost.

When it doesn’t matter anymore e.g the impact on user experience is irrelevant