The paper you mention is most likely BS, just like the document [1] they reference. They did their study with a bogus simulator. We had the opportunity to talk to those guys and after they admitted some problems with their simulator, they never retracted anything about those papers that are almost the only "research" hit that you get on STP. Very frustrating.
The only thing you need to be careful about is that your BPDU does not age out. The age is incremented by one at each step. The diameter recommended in the IEEE spec is too conservative and correspond to the use of an old software bridging running STP. With RSTP, just increase max age, it will not have impact on the convergence time if all your links are p2p (and they should).
As with any distance vector protocol, RSTP could get into a "count to infinity" situation if the root disappears from the network. In that case, communication could be affected until the old stale information is removed from the network. A matter of second indeed. There should be no temporary loop of traffic, only the control information can loop around. This situation cannot occur if the root bridge is part of the ring.
We have customers using RSTP on access rings for metro ethernet. We get decent convergence time of the order of the second. It's hard to get an exact figure as it depends on the number of mac addresses, the number of vlan and the location of the failure compared to the location of the blocked port.
Regards,
Francois
