I assume they are doing it mainly to avoid distribution costs; they simply push a web-app update via a single, well-standardized channel allowing DRM and trusted remote code injection (to prevent pirates from running it), and they don't need to maintain multiple internal tools to do that with multiple attack vectors. So this IMO might be an experiment in this fashion; they are working closely with Google, probably assuming they can't compete with them so it's best if Google's brightest figure out proper ways, and they can focus on their core business.
How could it work for them? Most decent sized projects would promptly run out of available memory and have nothing to do about it. And before that they have to deal with the ever longer list of restrictions put on web applications because of the trust model.