Not necessarily, it depends on how the reaction scales. If the reaction does not scale linearly (as is claimed) you don’t necessarily have to get more efficient, you just have to up the power until the output curve has increased past the input scaling. How big that is is determined by the efficiency of the input device itself, but it isn’t a question of if it will ever happen.