Ah, good point, memory consumption might indeed be a factor here. If that's the case I'd expect higher variance in the measurements perhaps - looking for that in the raw data could be interesting.