It's true, that the situations in which you care about it are few. The most recent performance issue I've come up against is actually java's startup time, because for reasons unknown my client decided java AWS lambdas were the way to go, has quite a large, complex codebase built around it, and now has complaints about startup latency.

I have noticed some younger engineers on C++ projects doing some kooky things "because it's more optimal", and have taken time to explain to them that the micro-optimisations they're using impact readability, are going to be done by the compiler anyway if they're at all useful, and are likely to be several orders of magnitude less relevant than some lock somewhere, or a bit of IO, or whatever.