my point is that what the benchmark is testing is the performance of cache hits, rather than the actual performance of the merkle expansion