Hence the standard technique of running unit tests using a private sandbox database.

The problem with using record counts as a pass/fail is that it's often difficult to roll back the clock to determine what else was going on in the database while your tests were running. When it gets to be too much of a nuisance to figure out whether a test really failed, you risk having some people adopt a "Oh, yeah, it failed, but that always fails, so we're really O.K." attitude.