The performance of MinMax k-Means is studied on several datasets and we wish to investigate if indeed its relaxed objective limits the occurrence of large variance clusters and how effective the proposed method is in overcoming bad initializations and attaining good solutions more regularly than k-Means.