> combined with server-side content negotiation.
> A browser should use the 'type' attribute to build its Accept header for
> this resource.

Unfortunately, the world seems almost as though split into:

- those that don't know that server side content negotiation is possible;
- those who know but can't afford to pay their ISP enough to allow them
  to configure the server properly.

The only sites I know of that do do server side negotiation of anything
but User Agent (using procedural, rather than declarative code, in that
case) are Google, some versions of Microsft's Windows Update page and
some images on the W3C site.  None of these can effectively use 
q factors as the common browsers do not allow the users to configure
them, relying on positional fallback instead.

As a result, I think that client side negotiation by using nested objects
may continue for a long time (although the first category above is unlikely
to be aware of that option, either).

(The same basic reasons explain why meta http-equiv is so popular
and also the use of scripting mediated redirects (and the, often resulting,
blank pages).)