This reminds me a little bit of a lengthy blog article by some FAANG engineer about picking optimal chunk sizes for I/O where he did a lot of benchmarking with /dev/null and /dev/zero.