Regarding the reason for the pandalisation and recovery, is it plausible that spreading the content across more domains has been the key?
Panda seems to dislike too much too closer related content within a site - like the notorious 5 ways to pick your nose example. I have found that killing of sub sections for different types of widgets, actually improves my rankings across the board.
I think google maybe uses external link or link juice generally to decide how many widget pages you can have. Too few links trying to prop up too much closely related content (though not actually duplicate) sets off the Panda bomb.
Would this fit with your experiment? Because it occurs to me that you are carving up content from one site and spreading it across domains? Of course, as you build more and more up in those recipient domains, eventually you may find yourself tripping the "too much too closely related content not adequately supported by links or domain trust filter" (ie Panda!)
Does this fit?