The distinction of threads v processes is not something that necessarily influences the logic of our code. If I want to say
a()
b()
c()
I can do that with a thread or a process and it doesn't matter what a, b or c actually do. However, with the "IO must be asyc" stipulation, I have to know whether any of those functions is an IO call, because if it is, I have to write it in a completely different way, even if the logic I need is synchronous. I don't like that kind of interference with my logic.
Well, with very high probability, your logic is broken in the face of errors / malicious users then.
Node's async structure is just taking the async model of the web one step further, into processing the same request asynchronously if you need to wait for something else.
You might be master of doing this logic right, but I've audited tens (perhaps hundreds) of systems, and all the synchronous ones got it wrong.
(Disclaimer: I've never worked with Node. But I've been writing async servers in C and Python since 1999).
Threads on the other hand, are the often the wrong optimisation -- processes are the right one.