Statistical fluctuations are observed to profoundly influence the clustering behavior of granular material in a vibrated system consisting of two connected compartments. When the number of particles N is sufficiently large ( N approximately 300 is sufficient), the clustering follows the lines of a standard second-order phase transition and a mean-field description works. For smaller N , however, the enhanced influence of statistical fluctuations breaks the mean-field behavior. We quantitatively describe the competition between fluctuations and mean-field behavior (as a function of N ) using a dynamical flux model and molecular dynamics simulations.