I find test coverage a useful thing to track, but I can't find a good metric. "Coverage decreased by 10%" sounds bad. If you're deleting pointless code that has tests, it's good! Perhaps total untested lines of code is better?