Here's another example of operation at a higher level: http://subtextual.org/subtext2.html
I'm not sure how it would scale though, perhaps in a collapse/expand fashion similar to the current 'focus' mechanism, or some sort of level-of-detail aware zoom that progressively replaces chunks with higher level representations.
The author also notes on the main subtextual.org site about issues of mutable state, but it's definitely an interesting take on language design, and may (hopefully) not share the same death-by-layout-complexity fate most visual languages eventually suffer.