While these benchmarks run at roughly 1500 i/sec setting the threshold
to 1000 leaves some room for deviations (e.g. due to different DB
setups).