Not really. This is a hyper optimised microbenchmark. The actual work being done is so trivial that any "overhead" is really significant and you can get order of magnitude speedups by doing arcane things with caches, eliminating tiny inefficiencies and so on.
Complete speculation, but I think you'd find that's true for most problems. Very little CPU time is spent on the business logic, and the rest is overhead.
Not really. This is a hyper optimised microbenchmark. The actual work being done is so trivial that any "overhead" is really significant and you can get order of magnitude speedups by doing arcane things with caches, eliminating tiny inefficiencies and so on.