My thought is that it's two totally different paradigms.
When you use the jQuery/Stripes approach, and you could substitute virtually any other combination here (Prototype/Struts, Dojo/plain servlets, etc), you're still dealing in URLs. You're still essentially creating abstractions: those URLs are a layer of abstraction above some back-end code. Now, with some frameworks that abstraction may be very thin, with others it's thicker, but it's always there.
With DWR, you're dealing in remote procedural calls. You no longer think things like "well, url X maps to class Y,method Z". Instead, it's simply "I want to call method Z of class Y and get it's return value", and that return value may be a simple
String, or it maybe be a whole value object.
You're right I think to say that the example you showed isn't complex, and probably isn't any more complex than the DWR answer, at least in terms of code volume. Where I think DWR has the advantage is in conceptual simplicity. Assuming for the moment that you're talking about a Java-based back-end, why should you have to mentally flip a switch and think of something other than what you're used to doing, which is method calls on objects?
Aside from that, why would I want to write code to spit out JSON representing some Java object? Because somewhere, you're either going to have to write that code, or at best find some library that does it and make use of it. DWR gives you that for free essentially (it's not JSON it returns, but my point is, who care!

).
For me, I've done code similar to what you're talking about with jQuery. And it's by no means difficult! But I find the approach that DWR pushes you towards, that whole RPC approach, to be simpler and more appealing from an architectural standpoint. YMMV of course
