Tweak division and string conversion routines for better performance - sadly we're still way behind GMP on these (though better than libtommath).