It could be done entirely without external requests: the browser would be an amp interpreter, no need for the amp js as all.
The hardest part is that amp probably isn't restrictive enough: you can still use a very large subset of HTML and CSS.
Very interesting thought. Probably enough HTML for tracking though right?
I'd quite like an RSS+reader view browser. At a guess, RSS+reader view+AMP could cover pretty substantial amounts of existing content.