I like these, although I wonder about the value of speculating about a performance optimization while neither just doing it nor profiling.
Documenting it saves the poor dev doing profiling in the future a bit of effort so they can come up with the better solution that you failed to come up with when writing the code.
Often times code has to be written and committed and I don't have the time nor the brains to come up with a novel solution that solves a future performance issue that is not yet and is not expected to ever become a performance issue.
TL;DR: it's just a "TODO profile this".