BTW, what about the 1/3 of a page restriction on tuple size?
I think that truncation has to strictly guarantee that the resulting tuple will be no larger than the original. Otherwise, you get into trouble here. But, that's still not my main objection to using the null bitmap.