By performance, I mean "How fast will an optimized $X$ be" or "Is Algorithm $Y$ likely to give good hardware performance?".
Are questions on crypto performance on-topic?
Yes. There is even a tag for those. However:
"How fast will an optimized $X$ be"
This is not answerable. The only way you can answer is measure an optimized implementation, and even then it could be optimized further. And since performance is platform-dependent, the best way to answer is by the OP to measure on their setup (which is not necessarily useful for others).
"Is Algorithm $Y$ likely to give good hardware performance?"
This is better, but not very objective as stated. Comparing to another (significantly better than Z) would be a possible way to make it more objective. Or having some other clear definition for 'good', I suppose.
What would answerable, objective performance questions be like then?
For example, comparing the performance of an algorithm with different key/field/etc. sizes. Those can often be answered in general, without the particular platform having a significant effect.
Or for something more like your examples it could even look like: "Which MAC algorithm would be faster than X in software, while fulfilling these requirements..." Here the platform could be important information, or it might be answerable in general.