It never fails. When a test suite takes a long time to run, I always start trying to do too much between two runs, leading to entangled unrelated changes, large deltas, and difficulties getting things to work.