If the memory used is being different for some reason, then I don’t see what’s unreasonable about expecting the execution time to also differ by a couple of tenths of a second because of that. That’s also part of what I generally called ‘non-determinism’. If the same code, run multiple times on the same machine under the same environment, behaves differently, I call that non-determinism, and there’s really nothing that can be done about that. Not saying that there’s something wrong with your code - just trying to explain the few tenths of a second difference.