Well I guess I sorta(?) answered my own question.

Seems like all match to my eye, unless I missed one accidentally. That is WITHOUT hwmonitor running in the background. When it was running I got 3 mismatch residual out of 25. All I did was rebooting the PC. So basically yeah, if you ran at least 50 passes and they finish successfully without errors, the occasional 2-3 residual mismatches are probably caused by background processes minding their own business.
Also I've noticed that multithreading actually lowers my GFlops down to around 330-340, and (unproven, can't testify) supposedly greatly increases your chances for residual mismatch.