I think you're making the same mistake as many other modern benchmarks do: GPU compute != general purpose compute. A Pi Zero 2 is easily multiple times faster

I think you're making the same mistake as many other modern benchmarks do: GPU compute != general purpose compute. A Pi Zero 2 is easily multiple times faster than a Cray 1, but not 300 times, except for certain very specific workloads. For general code that does lots of math but isn't GPU accelerated, I bet the difference in performance is significantly smaller.