I've just always felt that if I have 100% coverage of each sub, I must be testing the return value. To boot, I always overlap tests so that it's often not just one test file that uses the return of a sub, so that if I break that return, I may end up with broken tests in several places, if not just several tests within a single file.