I like some features of NodeJS, particularly JQuerification, websocket compatibility via socket.io, view and css engines that I cannot use with JSP (and of course, asynchron
To me the plan sounds reasonable per se. But from my experience it is important that your team is strong enough to bear it. In this case I would not go this route unless there are at least two good devs, one for the back-end part and one for the front-end. Otherwise it's just too easy to get lost when dealing with so many frameworks / concepts, and nothing gets finished.
Besides that I'd take care to make the communication layer between back- and front-end easily testable, which would rule out socket connections. If your performance requirements allow it, I'd opt for a browser-explorable REST style interface. This would also make it possible to drop the "fancy" front-end with reduced effort later on, and implement something in JSP or whatever. Just in case it gets out of hand...