I don’t think that the tank losses metric is interesting (in these results). I think what is interesting is the different ratio of crew losses to tank losses. There doesn’t seem any very obvious reason why that should vary wildly over time given that the equipment, the enemy and (very broadly) the operations remain similar – unless someone can think of one. The sample size issue is then one of considering simply the losses – are the losses big enough to constitute a significant sample size.
As John pointed out above, the odd figures are for the US tank battalions – I don’t doubt he is right and that what they show is that over the course of the campaign, these battalions suffered a large number of battle casualties of personnel who weren’t crewing tanks at the time – alternatively, the tank losses are being recorded differently from the battle casualties.