Aren't benchmarks supposed to be more or less reproducible? So just publish the results and the data?