Here are some ideas:
A) The disulfides in 3.1 are known to be problematic. They’re being rebuilt from the ground up for 3.2.
It might not have worked if it used CYS residue types instead of CYD residues. The former retains the sulfur’s hydrogen; it will clash horribly even with the constraints and probably overrule the constraint.
C) It’s fairly likely that none of the loop code is written to respect disulfides in the loops anyway – the two loop closure algorithms are intolerant of that sort of secondary loop restraint (or at least, they aren’t written to know about it)
D) I would try regular relax mode instead of trying to loop model with a disulfide in the loop.