One of the interviewee questions could be "Based on the profiler output, how many times would this code execute in a typical day, to the nearest power of 10? What code will it replace?"

string.reverse() and string.equals() is plenty good enough for smaller magnitudes or all-new functionality. Premature optimization just means you think you can outguess the profiler.

So first you code the simplest thing that works. Then you start optimizing in the most heavily used sections. If you have to go past the pointer-based solution in C to assembly that employs hardware-specific performance tweaks, it's time to just end the interview and go pet your taco cat.