Still, things have been like this since 8.2 when we implemented multi-row VALUES, and nobody's noticed up to now. Maybe the right answer is to change the data structure in HEAD and decree that we won't fix it in back branches. I don't really like that answer though ...
The merit of "won't back-patch" is that at least you don't punish those who are being lazy (in a good sense) but generating values in subsequent lines that conform to the type specification of the first record. We already implicitly accept such behavior elsewhere - though probably with better validation - so keeping it here is defense-able. We dislike changing query plans in back branches and this really isn't that different.
The concern, especially since this can propagate to a CREATE TABLE AS, is whether there is some kind of fundamental storage risk being introduced that we do not want to have happen no matter how rare. /me feels deja-vu...
