A source of contention in workloads, that leads to lower than ideal throughput is when there are hot pages that grow to the memory_page_max size, but can't be evicted.

We track some statistics around this with the failed eviction of pages that exceeded the in-memory maximum and maximum page size at eviction. Those statistics don't give insight into how much application threads are slowed down due to the process of getting the page out of the critical path.

It might be interesting to add a timing statistic around the eviction in __wt_page_release_evict. It might also be interesting to know how much of that time comes through __wt_page_in_func vs __wt_page_release. I'm not sure if the distinction is interesting though...