A lot of our tests which implicitly shard collections will never split those collections since balancing is based on data size and most of our tests do not insert much data. It would improve our test coverage (both of balancing and other features which are impacted by number of chunks) to randomly split collections in these tests as well.

We could achieve this by augmenting the "implicitly_shard_accessed_collections.js" helper to also split these collections. However, we should also make sure that we implicitly split internal sharded collection such as the sessions collection so that we have this coverage for those collections too.