I've seen that before when the large range features were the most important for the clusters I wanted. It was essentially doing feature weighting but it was implicit in the scales