All I was pointing out is that worst-case latency on a throughput-optimized OS can be unbounded, and it's not totally unexpected to see 1ms+ delays between those two lines of code, even if the clock implementation is flawless. Even in a RTOS you have jitter from cache misses and whatnot, just to a lesser degree. Code sensitive to clocks not being correct is often sensitive no matter how small the deviance is, and smaller deviations simply make bugs less frequent. Treating that point-in-time estimate as anything other than an estimate from the recent past can lead to code that looks flawless but occasionally breaks.