Replacing the explicit mod with another call to hash() saves a little time because the int hash doesn't need to do division at all. Because optimizations already done put a small bound on the bit width of the integer to be reduced, this saves just a little time regardless of how large that int is. In percentage terms, it's most significant for small fractions, where the final division accounts for the largest percentage of total runtime. For Fraction(1243, 5813):

before: 200000 loops, best of 11: 1.92 usec per loop
   after: 200000 loops, best of 11: 1.78 usec per loop