I don't think there's much need for heuristics like "rate of effectiveness change times donation must be much smaller - say, a few percent of - effectiveness itself."

If you're really using a Landsburg-style calculation to decide where to donate to, you've already estimated the effectiveness of the second-most effective charity, so you can just say that effectiveness drop must be no greater than the corresponding difference.