I'm indeed fairly confident that a missing `font-variant` doesn't trip a fallback codepath in CRe (as opposed to a missing glyph).
Our resident CRe wizard @poire-z will probably chime it at one point to say how feasible something like that would be

.
(Or, more generically, if it even makes sense and/or what the actual specs say about the canonical behavior (if there is one) in such an instance

).