Anyone here know the nitty gritty details of the segwit discount?

As I understand it, an old node sees a segwit transaction as an always true transaction…and that the witness data doesn’t affect transaction validation for old nodes. Do the old nodes still download this data?

I know there is a limit on the size of segwit blocks (hence the discount on the witness data), but is there any limit to the size of other always true transactions with segregated witness data not conforming to the segwit format?

In other words, is the arbitrary size loophole close?