It's probably the same tech as the kindle cloud acceleration tech. Basically the server (cloud) accelerates the page by fetching all dependent objects in parall

It's probably the same tech as the kindle cloud acceleration tech. Basically the server (cloud) accelerates the page by fetching all dependent objects in parallel on th faster network, then send the result over a single compressed channel. This saves the chatty nature of http and if modified since requests and of course the startup/tear down times per origin server.