If the result sets are that large, you have enough time to generate a program and compile it, though as you say a more streamlined approach is to use a JIT library. Postgres uses LLVM for that, and depending on the workload, maybe you could even benefit from using a GPU. None of this is new.