We haven't used the flakiness for builds, but just for performance tests. So we set up our service to point to a flaky MJ and the use siege to see how it handles it.

In one instance saw how certain scenarios meant we weren't releasing connections which then killed the app.