The Mixed10KThread template for a few perf workloads has a very subtle but important nuance that makes it hard to interpret performance results: the read queries match on an increasing set of documents over time because we insert more data that matches the query predicate over time. Read throughput is directly influenced by write throughput, because because the more data we insert, the slower the queries get.

We should consider imposing a limit of 1 on these queries to isolate these effects, as interpreting results can be incredibly misleading.