Jonathan Rosenberg (Bell Labs) presented further results on RTCP timer reconsideration (draft-ietf-avt-reconsider-00.ps). The plateau effect which has been observed is due to a large number of receivers joining at the exact same time. If the join is spread over a short time (instead of an instant) the plateau goes away. The problem was mainly an effet of the simulation, since in real life an exact same time join is not possible.
A further problem has been noticed when a large number of receivers leave a session simultaneously. This will cause a flood of RTCP BYE packets, which may cause congestion. In addition, it will cause the group size estimate to be incorrect in those receivers which remain, since they will time-out existing receivers (which haven't send an RTCP packet, due to having a large reporting interval). The solution to this problem is to do reverse reconsideration, whereby the reporting interval is recalculated when BYE packets are received.