Testing is not a boolean. You won't notice if some item on a checklist took a couple of seconds during your daily device test, but a user will notice when that action is all he's doing, over and over.

You say "testing is not a boolean", then imply that testing "over and over" is? Developing on the emulator doesn't stop you from doing nice, lengthy real usage testing on the devices once you've got everything in decent shape.

That's true, and there's no reason most people can't use their gym memberships, but statistically they won't. Nobody's saying that you must do all your testing on average hardware — just that they prefer it because it forces them to put proper time into it.