Steady-state performance of incremental learning over distributed networks for non-Gaussian data