(Discussing TCP vs. wedge approaches)
g) a wedge implementation will have interesting interplay with transport layer RTT and throughput estimates. OTOH, that information is actually more host-to-host (or path related) in nature than connection related.
You'd think so, but I sure haven't heard of a lot of people doing ftp://ftp.rfc-editor.org/in-notes/rfc2140.txt in production TCPs...
I haven't really checked the BSD stack code recently, but it does keep the RTT and throughput estimates in routing table entries and not in tcb entries. In fact, it has a cloned routing table entry for each active host, and that entry carries the RTT and throughput estimates over consequent TCP connections.
I would need to study to source code more closely to be able to say anything more.
Maybe the wedge approach would provide an incentive here?