Comment by emfree

I think the author is specifically evaluating low-concurrency in-memory workloads here. The previous post describes why regressions for those workloads might be "not a surprise": https://smalldatum.blogspot.com/2017/05/the-history-of-low-c...