It might be stupid, or it also might just be naive, or a shift in priorities.

100x throughput improvement might just come from caching results from earlier computations (less naive) - at the cost of 10x memory footprint possibly (different priorities).