You could be right, maybe there's a hormetic valley somewhere in the risk curve. I think it still has to be linear close to the background radiation level, and grow faster than linear as one approaches the lethal level, though.

Isn't it true that the proportionality coefficient for LNT was calculated using cancer rates for atomic bomb survivors? The dose for them would be very rapid, and there would be little time for enzymes to be produced. This would explain why such a valley didn't show up in that particular set of data.

LNT has been checked as far down as the statistics lets it be checked (including with animal experiments). Those doses are much higher than the doses I was talking about, the doses that actually matter for nuclear accidents. It's conceivable that the slope of the effect/dose curve at very low doses could be very steep. The results at those higher doses wouldn't be able to tell.