I like this characterization from James Robertson. It really does go back to before Alan Kay conceived of Smalltalk, Kay observed several distinct things behaving similarly, like the Burroughs B5000 and biological cells. And now HTTP is just like that. HTTP is a basic message passing mechanism that can be used in all kinds of situations. It's not ideal but then what is? HTTP is wide spread and we're seeing it spread even further than current conceptions. Languages are becoming more dynamic, and so is messaging on all kinds of networks. Spinning up a new process for passing messages via HTTP and XMPP has to become easier still.
Smalltalk arrived on the message passing frontier a long, long time ago. In a lot of ways, HTTP messaging resembles what happens in Smalltalk - you send the server a message, and if it doesn't understand, it sends you back an appropriate HTTP error message (kind of like a DNU in Smalltalk). The server doesn't crash, it doesn't throw up its hands and stop; rather, it awaits the next message.
This kind of architecture has to be flexible, and growable at runtime. Smalltalk has been that way since the beginning, and HTTP servers operate in much the same way - you can add messages that they'll understand in well understood, dynamic ways. It's kind of nice to see people understanding this strength :)