I've long been interested in bivariate versions of the Zipf distribution (or similar head/long tailed) distributions.
If you found your way here, you might be interested too. If so,drop me a line, especially if you have data to share on this.
Around 2008, there were no web pages with 'bivariate Zipf' in them. So I put this one up. Now it's mid-2011 and I find 9 pages via Google. A paper with Justin Dyer on this topic just got accepted at EJS.