I'm inclined to think that it's not really necessary to worry about
invalidating a per-connection cache of "is this function safe to ship"
determinations.
So: yes to a local cache of all forwardable functions/ops, populated in full the first time through (does that speak maybe to using a binary search on a sorted list instead of a hash, since I only pay the sort price once and am not doing any insertions?). And then we just hold it until the connection goes away.
Yes?
P.