The robots could then visit the page, get an HTTP 200, and determine the correct url to index from the canonical tag, while users would get transferred to the new url by their browser.

Do you think this would work?

It is interesting that Google are now allowing this cross-domain, and I think there definite SEO benefits as you describe. However, I would question whether it’s use would ever be optimal across domains, as it would seem far better from a usability point of view to present a single url to a single resource, accomplished with a 301 redirect.

In your example, I’m not sure who Simon Baker is, or why his domain is a CNAME alias of news.bbc.co.uk, but it would seem confusing for a user to access bbc content through that url, rather than be redirected to the proper domain. Or if it was a testing site, it should not have been indexed in the first place.

So, I’m in total agreement about the usefulness of the cross-domain canonical tag where the webmaster cannot implement a server-side redirect, but it would seem second best in all other situations.

Conversely, it would seem to be an eminently sensible solution within domain, e.g. with your affiliate id (http://www.example.com/example-page.php?affiliate=blueberry) or on a page with mutiple sort options (http://www.example.com/widgets?sort=asc), where the difference in layout necessitates a separate URL.