Currently the massive-scale testing is in one test suite that fails when the test isn't run against MongoDB with the GroupCollections change.

We should split this test into 3 parts tuned accordingly so that all pass

1. MongoDB running the largescale test with GroupCollections set
2. MongoDB running the largescale test without GroupCollections set
3. MongoDB running the LargeCache test (WT-3352)

These three tests should then be able to pass consistently and produce the "high watermark" they used to pass, which can then be added to a plot and used for trending.