Spelling out the numerator/denominator calculation in the __mod__ special method, and actually implementing __divmod__, speeds up both operations by 2-3x. This is due to avoiding repeated Fraction instantiation and normalisation, as well as less arithmetic operations.

$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'a%b'
50000 loops, best of 5: 9.53 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'a%3'
50000 loops, best of 5: 6.61 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'divmod(a, b)'
20000 loops, best of 5: 14.1 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'divmod(a, 3)'
20000 loops, best of 5: 10.2 usec per loop

$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'a%b'          
100000 loops, best of 5: 2.96 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'a%3'
100000 loops, best of 5: 2.78 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'divmod(a, b)'
100000 loops, best of 5: 3.93 usec per loop
$ ./python -m timeit -s 'from fractions import Fraction as F; a = F(-7, 3); b = F(3, 2)' 'divmod(a, 3)'
50000 loops, best of 5: 3.82 usec per loop