Websockets and WebRTC really are not enough at all, and this is a very widely held misapprehension. Websockets (and their HTTP/2 equivalent) are just a dumb pipe for shifting data - just like a TCP socket, but tunnelled over HTTP. Meanwhile WebRTC defines how to encode and transport real-time media streams over the 'net... but deliberately avoids specifying how to discover or signal the existence of those streams in the first place. In analog terms WebRTC is like a phone line without a phone exchange.

I agree that the open Web avoids silos through URLs and ad hoc HTTP APIs. But this is insufficient for a hypothetical VR universe of immersive services: you don't want the only way for a user to be able to navigate between services to be to 'click a link'. And it's arguably a failure that almost every single web service ends up publishing its data via an incompatible ad hoc HTTP APIs - for instance, given Facebook, Slack, HipChat and Skype all provide supersets of precisely the same basic functionality, how come each has its own proprietary custom HTTP API to do the same thing? Wouldn't it be better if there was a standard open data fabric into which they exposed their data?

The idea here is that VR will suck if it's just a bunch of disconnected apps, and instead you need decentralised primitives for sharing Identity, Avatar properties (e.g. position & orientation), shared VoIP, shared Video, even shared world geometry/assets/physics in order to stitch it together into a coherent experience; both for users or developers.

The web was successful because HTML is so simple, forgiving and flexible - and you end up with essentially the minimum successful viable hypertext system. One could probably limp along with a VR Web which is just a bunch of webapps which happen to have chunks of WebVR UI. But it could be so much more, while still being flexible and forgiving and lightweight. At least this is Matrix's hope :)