This is really neat, great work! I wonder, do people perform the kind of edits that this makes efficient very often?
For example, in the video it was predetermined exactly what was going to be moving and where it was going to go to. From my observations, these sorts of large scale changes are rarely handled so easily. Context helps people map things cleanly, and the contextual change of zooming in and out forces you to constantly rebuild that map. It's an issue I've seen over and over again with any zoom-oriented interface.
Regardless, this is cool work. We need new and interesting interfaces for code - I'm hoping to show some fun things myself at the Live Programming Workshop [1] at ICSE[2].
[1]: http://liveprogramming.github.com/2013/ [2]: http://2013.icse-conferences.org/