Thanks for your contribution!  I myself have been thinking about a DS-like approach to the multinomial inference problem and your paper is relevant to this.  I'm definitely interested in this and, after I have a chance to work through some details, I'll post here a formal review.