Also, API endpoints for friendship might have a follower_since date, and you can approximate the history. In so far as unfollowing is rare compared to following, it would be accurate.
Unfollowing is not as rare as you think. Out of all follows/unfollows SocialGrapple has tracked so far (for detailed accounts), 35% are unfollows of the subject account (63% unfollows from the subject account).
I did check the API, Twitter doesn't give you historical relationship data like that. This is something that SocialGrapple was built to solve.
Scraping is something I haven't considered, though being a tertiary source of data is something I'd like to avoid (I already have nightmares of Twitter shutting me down for no reason). I'm not even sure how accurate TwitterCounter's data is, at least I can account for my own and Twitter's mistakes. I'll give it some more thought nonetheless.
We doubled followers in a weekend by testing but not launching http://dropbox.com/free
Then we did Dropquest, and got another bump.
If we add twitter following to the getting-started quest http://dropbox.com/gs we could blow those changes out of the water.
A graph with context: https://www.dropbox.com/s/mwmyb8zmt4kufis/at_dropbox.png