Not only do you need non-blocking code for things like database access, but maybe more subtly, you need to be non-blocking for anything which is potentially CPU heavy. Case in point: encoding a payload in say xml or json. If you have a response which are big enough that they need say 20 ms to encode, all response whose processing have started but not yet finished will see a 20 ms additional delay. Now, you can incrementally encode into json by inserting new "scheduling points" at regular intervals, but doing this for say zlib compression, not so easy.
Not that I have a lot of experience, but IMO, async code should be reserved to the least amount of code possible, because everything becomes a bottleneck. And good luck trying to profile your async application with all those callbacks when you have performance issues...
http://en.wikipedia.org/wiki/Comparison_of_server-side_JavaS...
which I will probably work through at some point, I'd really like to use JavaScript on the server but I'm not sure if node.js is a good fit.
That is my understanding as well - because a long running task will block everything, you have to be MORE careful in node.js than in a traditional threaded environment. I actually think that this model works really well in 10% of applications, and a threaded model is a better choice in 90% of applications.
I also believe 'less mature developers' see, 'wow, server side javascript', get excited by that, and don't put enough attention on the more important architectural differences of node.js