I don't think you can ignore memory access. The whole "hyperthreading considered harmful" was because of shared cache between contexts. That's why I think a proof is in reach without cache, much harder with one...