The finished program went down in runtime from 533.7s on my machine, to 24.42 seconds. If we adjust for machine speed (since the author's Python ran the python code in 474s), the comparative optimized result would be 21.7 seconds compared to author's hand-optimized lisp code at 30.2.
Note that this is probably an ideal case for Shedskin (which is a relatively new product: a 1-man job that's just a few years old compared to, I guess, hundreds if not thousands of man-years of research that have gone in LISP implementations).
Still, an interesting result.