I guess that when the problem is to really scale a lot the big cost may be the energy not the hardware, so the performance the coder can bring you for watt is crucial.
And to use few CPU cycles you need a programmer that is smart enough to know both computer science to select the best algorithms and data structures and low level programming languages to lower the constant times involved in this algorithms.
And to use few CPU cycles you need a programmer that is smart enough to know both computer science to select the best algorithms and data structures and low level programming languages to lower the constant times involved in this algorithms.