That's only true if your utility function is linear. If your utility function is nonlinear, and you care about humanity existing, but you don't care as much about how much humanity exists, then a doomsday device isn't nearly as bad if you know humanity will continue to exist in a parallel universe. I assumed that this is why someone would prefer 90% of the measure of the universe being destroyed to a 90% chance of the whole thing being destroyed. Is there another reason you'd prefer the former?