I think approaches like this are fundamentally flawed: you're spending a lot of effort to shoe-horn an interaction into control structures that it takes a Ph.D. to understand and that are essentially black boxes.
Most of computer science is based around the principle of abstraction... Google for Judy Trees/Arrays. A Judy Tree is basically a glorified hashtable implementation that's designed to be maximally efficient independent of local memory architecture ("cache oblivious" algorithm)... and I believe it's around 20k lines of code. Extremely complex implementation, but a very easy black-box interface to understand ("everyone" knows what a hashtable is).
If someone can design an *effective* abstraction on top of HTTP that makes it behave like a single thread of execution, then great! It will be that much easier for the average developer to design an effective web application. The devil's in the details, though; an imperfect abstraction is going to break down too easily and end up just sucking. But don't discount an idea just because it's difficult to understand/implement (otherwise, we probably wouldn't have most modern programming languages in the first place).
--