The problem is in the interaction. If you imagine the "normal" X chromosome (let's call it N), and the "bad" X chromosome (B), then a network of entirely B cells or entirely N cells works fine, since they're all on a standard "protocol" if you will.

However, if you undergo X inactivation in an individual that has BN genetics (one copy of each), then you get some clusters of cells with each type - And at the interface between B and N you get weird, glitchy behavior that can cause symptoms like seizure because they aren't quite fully compatible with one another.

At least, assuming I understood correctly.