(1) You can use XMLHttpRequest to send a HEAD request. Encode the RPC parameters in the URL. Encode the results in "X"-prefixed headers of the 200. No body necessary.
(2) As far as "getting the document controller to stay online", this ain't necessary if the JID referred to in the namespaced element of the loaded document refers to a Pubsub node. The controller can be knocked offline, the node still exists. This is a better model, but server<->client IQ and messaging, and client<->client messaging and RPC would all have to be re-worked around this new model. Bring on FOSDEM.
One of the things that's struck me with lurking on p2p-hackers and reading about DHT's etc. is that as well as building monolithic filesharing apps like eMule, we're also seeing p2p primitives appear, which can be bolted together just like cryptographic primitives to produce systems like Kenosis. This is the style of BT in doing one thing and doing it well.
(3) f you retrospectively change the title of your blog post, permalinks break. This is annoying.
(Also: I can't find a parser generator that will output a parser in ECMAScript. This probably proves that I'm crazy for wanting such a thing, but that's OK. I know I'm crazy.)