I'd disagree, based on my observations in both environments. Law of large numbers shows that it is easier to hide unusual behavior when number of samples is large and the observer is focused on aggregate-based performance measurement.