You need to compare the variations compared to the expected variation for a Poisson distribution with the observed rate to judge whether or not they are significant or likely arise from noise. Just looking at the deviations in absolute number isn't telling you anything.