Looks like you forgot to actually use the use_calloc parameter you put in the long_alloc prototype. It accepts it, but it's never used or passed to another function. So right now, the only difference in behavior is that there is an extra layer of function calls involved in _PyLong_New.

On that note, might it make sense to change all other calls to _PyLong_New within longobject.c to use long_alloc(size, 0) to at least make long object's themselves avoid the overhead of the extra function call? Maybe I'm overestimating how much of a difference an extra C level function call will make, but it seems like PyLongs make new PyLongs an awful lot.