Yeah, but Peter makes the case that people want it for global experimentation. "We think we can safely drop this humungous index that would take us days to rebuild, but before we do let's make it invisible and run for a few days just to make sure." I guess we could do that with a GUC, but it seems ugly.
On that front what's the proposed behavior for cached plans using said index?
IIUC with a GUC you'd have to force clients to establish new sessions if you wanted all queries to be affected by the new setting whereas using cache invalidation you can affect existing sessions with a catalog update.