Variance regression models in experiments with few replications