I think you would use the hypergeometric distribution to assess the significance of overlap between two experiments, and then for more experiments would probably multiply the p-values (sort of like adding in more dice to a dice role). Since every array has the same 10,000 gene universe, you can calculate the significance of drawing the same set of genes between arrays in a given sampling (i.e. the set of genes called at a 2-fold cutoff). This question is very similar to one that was asked before. (though some people were confused about whether it was one experiment with 5 replicates, or 5 experiments, but like your question, it was actually 5 experiments).