Like Alcaro said, you need to focus on some kind of benchmark (even just one) that directly represents a user journey, not microbenchmarks. For example, if you are experiencing slowness or jitter in some game, then you can measure frame time and/or standard deviation and see if you can make a statistically significant difference. FWIW, I don't think we have evidence that suggest noticeable slowness or jitters are coming from pthread_mutexes.