I know why it exists, and my point is they are not trying to show a gazillion txn/sec on 1MB of data. You are looking at a dataset that is several TB's. They have far surpassed the point where a vendor is trying to cheat by putting all the data in L1 cache and claiming to be fast.