Various Notes
First, I'm gonna be at OSCON the same week that TAE is going down. If you were counting on me buying you beer while in SF, you'll have to collect from Dylan (he's good for it) or one of the other Dojo's.
Next, can y'all please stop conflating Comet with a particular HTTP-level mechanism for achieving the stated user interaction goal? It's getting quite tiresome to hear folks say things like "long polling or Comet" as though they're different. The XMPP HTTP binding guys even go to great lengths to explain how their Comet technique (BOSH, aka: "long-polling") isn't Comet. Almost as entertaining as it is wrong. Long polling along with other techniques are ways of implementing the basic Comet pattern. The general description of the pattern contains no preference for one or the other. It only requires that you not naively poll N seconds.
Lastly, I'd like to express my disappointment that Jack didn't include Dean's excellent Base2 in his latest round of benchmarks. It's doing us all a disservice that he's not including the fastest engine around in his benchmarks.