Poker-AI.org

I'm matching my different versions of my bots each other through an acpc server.
What is the number of hands I should run them? Is there a paper on the math behind the calculation of number of hands, which could be called statistically significant? (I guess its just basic statistic, but what should I look up for?)

I know, a with duplicate matches I could reach faster a confident result, but I don't plan to change the dealer code in the near future. (If there is an apcp server with the feature to run duplicate matches, please let me know)