What is the web anymore?
There was a time when the Web was about setting up a way of sharing content, via hyperlinked documents, that could be rendered somewhat consistently on any platform.
It almost seems like the proliferation of webservices and increasing levels of functionality (complexity) being wedged into the browser are as much of a ploy to get around the network firewall security nazis as it is to usurp OS's and application platforms. In the old days, if I wanted a client server app to pass data over the Internet, I would establish a protocol, secure it with some sort of encryption, reserve a port and let my networking guys know that the port needed to be mapped to a server.
Today, a request to open a new port is likely to be met by the security guys with more resistance than permission to have a night out with their women. But now, as long as I can figure out a way to render a half-usable UI using Javascript and DOM, I can move about all the data I want using XML or JSON, and it is met with far less scrutiny by the security folks. I'm not convinced that this is a good thing for data, security, or the efficiency of Internet. It would make no sense to implement things like SFTP or Usenet as webservices, but that's about what we are doing with almost all Internet applications these days.
Getting back on point - if Google wants to make an application platform, they should start from a clean slate and quit encumbering the web. HTML5 has some elements of this (sockets, media rendering) but it's not enough, it's creaky, inelegant and has a lot of the traits of something that has been kludged by committee over time.
Developers need a real, distributed application platform. Stateful connections (without cookie lameness), straightforward persistence, consistent and known media formats, data interchange without having to muck about using XML or JSON, scalable vector graphics, a consistent UI toolkit. All this sort of sounds like what Java could have been, but the execution was a failure. Google can make this happen. They can build runtimes for Windows, Mac and Linux. Better yet, Google could create its own OS to optimize the implementation (but it should be more than Chrome running as a shell on top of Linux). This could finally be the thin-client solution shops have been promised for time-immemorial.
Going forward, keep the new releases well-versioned. Major new functionality (touch support, speech recognition, 3D, new media formats) should be major releases; minor functionality and bug fixes should be minor.
The native code browser add-in approach described seems like the same bad idea of making a browser and the web do things they were not designed to do, with the historical precedence of Netscape plug-in's, ActiveX and Flash. Start over.