It's not the allocation itself that's faster, it's faster because it avoids the overhead of crossing the C++ -> JS barrier