Just to quantify the improvement:

Before:

$ ./python -m timeit -s"x='abc'" "x[::-1]"
1000000 loops, best of 3: 0.305 usec per loop
$ ./python -O -m timeit -s"x='abc'" "x[::-1]"
1000000 loops, best of 3: 0.275 usec per loop

After:

$ ./python -m timeit -s"x='abc'" "x[::-1]"
1000000 loops, best of 3: 0.262 usec per loop
$ ./python -O -m timeit -s"x='abc'" "x[::-1]"
1000000 loops, best of 3: 0.253 usec per loop

For some reason, when I run pybench, the timings vary from run to run so
much that I cannot even tell the difference.  (Run to run differences
are larger than patched to original.)

FWIW, the micro-benchmark above shows 8% improvement with "-O" and 14%
improvement without.