I listened to a talk once where they said that this is sort of how AV1 compression works, where there are many possible ways to compress a block, and you can choose the best one at runtime by computing each and seeing which is best.