The non-browser browser

Consider the number of websites that use a shit ton of styles, images, advertising, javashit, whatever. All that is shitty cruft that you don't want to have to wade through to get to the important part: data.

Fortunately, many major Web 3.0 Bullshyte websites have RESTful APIs too (mostly JSON), allowing you to gain access to structured information in a consistent way.

Enter stage left: A scriptable, customisable program that, given an interface description for a particular website or data source, will show you the data you want in the format you want.
A true separation of interface from implementation.

However, there are drawbacks. Each data source will need its own interface design (but enough people who care coupled with a few public databases, that will not be a problem). Websites will lose ad revenue. The WWW as we use it will change.

So, /anus/ - is it worth it? Is there a need for such a program? Are there more, more serious drawbacks I haven't thought of? Am I just overreacting to Youtube's fiftieth redesign in the past three months?

Eventually you will face the problem of certain websites simply not working.
And nothing of value was lost.

The "sensible subset" is difficult to determine; some things like CSS animations, gradients, rounded corners, etc. are obviously eyecandy, but what about other heavy but possibly useful features like embedded SVG/MathML? It's tempting to start with HTML4/CSS2 and add to it, but then you'll eventually end up with something as bloated as the full set.

This is an interesting read:
http://www2012.wwwconference.org/proceedings/proceedings/p41.pdf
AOL and Picasa both contain large images but the CSS energy consumption for AOL is much lower than Picasa's. The reason is that AOL uses HTML tables to position its images while Picasa uses CSS to position images. This nicely illustrates how positioning using CSS is less energy efficient than positioning using simple HTML tags.
Surprised? I sure wasn't. All that "but it isn't semantic" bullshit can't beat actual measurements.