There's no observed instance of such "earlier repair," it's just an invented "hypothesis." Only the non-linearity is observed, and not even by majority of the experiments. The explanation for the observed non-linearity is that the single impacted cells simply have time to die (or be targeted by the white cells) at low levels, whereas at higher dosages there are enough cells for cancer to progress more often. It's just the game of probabilities at the levels at which we have enough noise (low enough doses).