I thought there was general agreement that a proportional limit was better than a simple number. One disadvantage to a simple per-package limit is that you can defeat it by splitting something up into more packages. A proportional limit seems more sensible to me.
Also, I think it should be a condition of any proposal that it recognizes the presumptive validity of our past practice--or, it explicitly say "this proposal is intended to refer to new software, not to things already in the archive". If it doesn't have a grandfather clause, then the alternative is to go through all known-to-be-affected packages and make sure they meet the test.
