Because we run druntime tests concurrently (both debug and release), it's possible to be running the coverage tests concurrently in both debug and release. However, the coverage tests output the same list file, so there is a potential race which results in failures during test.
This causes both the master and pull requests to fail randomly.