At Typewire we too started on Node and moved to Java (using Netty for NIO). I really truly love JS and wanted it to work, but the throughput on Node for open connections just wasn't where we needed it to be.
The best use case I can come up with is using node, socket.io, and knockout to quickly build a shared-state application (so long as there is a way to reconcile communication latency, which is why a declarative library like knockout is a good thing). I'm trying to come up with a good problem to solve with that tech stack for this year's Node Knockout.
In short, I agree, Node lends itself nicely to a particular subset of problems, but a lot of the places where it is presumed to shine (speed/real-time/code reuse) others are moving in to follow its lead.
