I meant "mining becomes more efficient" in the sense that using it as a heater is an efficiency gain over not using it as a heater. I didn't mean to imply that legacy non-heater miners also become more efficient.

Intuitively, you're right. However, I read an article (can't find it now, still looking) that presented a convincing case that if there is $X up for grabs for doing a certain amount of work (mining), then the amount of work that is done increases until $X is being spent on the work. So if you're getting heating out of it as well, X increases, but that just means more mining is done until it's no longer marginally profitable.

But intuitively you're right. So I don't quite know where the mistake lies.