Drivers run tests on different cluster configurations, and all of these cluster configurations are run on the same machine (the one the driver developer uses). The tests deal with very small amounts of data, several megabytes at most for the entire test suite. Even so my mongod processes are growing from 150 mb RSS to 500 mb RSS.

The commonly referenced tunable is wired tiger cache size, and this is specified in gigabytes. 1 gb is too much for driver tests, and many other tests in fact. I would prefer to have a cache on the order of 100-250 mb.