Logged In: YES 
user_id=973611

Here's the 3rd part of the patch (long_mont.diff; Montgomery
Reduction), diff'd against 2.4a3 and cleaned up a bit.

Note that this doesn't include negative exponent handling. 
If this patch is accepted, I'll make a new tracker item for
that, since it's not an optimization, just an "opportunistic
feature" (it builds on one of the helper functions needed
for Montgomery).