It wouldn't be a global list. It'd be unique per instance, and each instance would manually have to add their preferred lists with silenced users.
Starting a new instance would create new users, unknown to the existing lists, which would manually have to be added to any site's trusted lists before a blocking would be in effect.

What I'm thinking is that if the network works "as it should", i.e. small instances where all users know or at least has good knowledge of who the admin is, everything would be spread out well enough anyway. (and a simple "no direct mentions from unsubscribed remote users" user-configurable option would be enough).