It's not being scuttled by randomness, since the times are so wildly different; my measurements of my Rust code indicate "only" 20% of the time is being spend i

It's not being scuttled by randomness, since the times are so wildly different; my measurements of my Rust code indicate "only" 20% of the time is being spend in the RNG ranging up to 35% if I use StdRng rather than XorShiftRng, this difference is peanuts compared to the 750× (not percent) speed-ups the Fortran/Rust/C/C++ sees over the original Python (and even compared to the 30× speed-up seen over the optimised Python).