If you invent a hash function that is two times more computationally expensive, then you will simply drop the hash rate by half.
Speed is the goal of the miner, while lag is the goal of the developer. The idea is for the hash function to be a lot less computationally expensive. Anything that keeps the processor paused or preoccupied with low-power, time-consuming overhead, like context switching, interrupt requests, memory access, searching arrays etc. But the overhead must be required for valid hash outcomes, so it can't be skipped or optimized.