i think we agree that per cpu caching seems superior. i’m looking for the other side of this. most allocators seem to have stuck with per thread.