Clause 9 currently says that when a node passes the token, it waits for NS to begin using it. If no such usage is seen, the node may retry ONCE, after which it uses PFM to find a new NS.
This seems a bit severe: most nodes will do 3 or more retries at the APDU level - not just one.
Once a node has been dropped from the rotation, it is locked out until it is offered a PFM. The PFM poll may not come for 50 rotations. I have seen links with large max-info-frames where rotation time is 5 seconds or more, so it could be a looong time.
If the dropped node uses postponed replies, either always or as part of transferring a segmented reply, it can easily become overwhelmed - other nodes continue to make requests of it, but responses must be queued until it can send previously queued responses.
It would be trivial in both the standard and most implementations to increase the token retry count.
Is this worth a formal proposal, or have the Olympians already considered this and discarded the idea?