How deep does the parallelization rabbit hole go?