It’s also a multidimensional problem, so sometimes people improve the state of the art in one dimension, but either at the expense or the neglect of another.

I think it’s a bit like any applied science. There’s a theoretical limit to the efficiency of an internal combustion engine. We grow closer to it, often with more complicated machinery, occasionally with better construction (ie, the manufacturing becomes more complex so the product is less complex). And we balance the consequences (pollution, reliability, cost) against the benefits.

Once in a while we substitute a different model (E.g., Atkinson) with small but important differences in the theoretical limits, by looking at the problem differently.