I think the question is whether Knot Resolver follows the letter of the RFC, like BIND, or whether it is less strict.

I'm not aware of RFCs saying that resolvers should fail in such situations.  My understanding is more like it's allowed not to work.

Anyway, it would be better to continue this thread on knot-resolver-users@lists.nic.cz
(I already have posted there about this a couple days ago.)

--Vladimir