Running under VMware skewed the first results. We have been hustling to get the code to version 0.1 and I didn't get a comparable test PC in time. We will be rerunning all the benchmarks soon without VMware.

As for using node.js, we're starting to test various communication channels -- websocket versus long poll techniques versus straight ajax etc -- and it's faster for us to be working in a single project as we bootstrap up the tech curve.