Since the merge of the dynamic compression change in WT-4433, all of the long tests have seen improvement. The long read-only test (test 4) shows almost 4x improvement in throughput sometimes.

However, the throughput is now wildly variable, up to 100%. It seems to have a low of about 42M operations and a high of 80M operations but run-to-run, it changes wildly between runs.

I have attached the plot. The old code was a solid straight line (the bump on the left were a few earlier runs of a dynamic compression change). We should investigate to see what is going well/poorly when the test varies by 100%.