Or canvas substitute. SVG/VML is the technology that actually augments HTML documents. Canvas is just super-slow Flash/Silverlight-like pixel renderer, a “box on a page”, as you said.
There is a barrier between a Flash/Silverlight app and the remainder of the page, that's not true for the JS code driving a canvas element, it's the same code that drives all of the interactivity on the page. This makes a difference.
Why, you can interact with Flash/Silverlight embedded object as well. (A perfect example is using Flash to fake Websockets) And both support Javascript. There's really not much difference.
This is why there's so terribly many flash/silverlight apps driven by external javascript? And why, on average, there's so very much cohesion between flash/silverlight objects and the pages they live with in?
In practice, the wall clearly exists. The average flash/silverlight app is an alien box sitting amidst its host page, with very few and generally comparatively minor exceptions.
You claim that less than expected (by you?) people use available technology in such and such way and thus that imply that there's something wrong with technology?
I would suggest there's really not so much use cases for Flash/Silverlight—HTML in-page communication. Or maybe some people just don't know their tools very much? None of these would be solved by creating another similar technology.
EDIT: Just look at the new HN top story for an example of JS-Flash interaction: http://feross.net/instant/ :)