If you were to cluster with all the k from 1 to 12, then how would you decide which was the "best" number of clusters?
Any deterministic clustering will always fit better with more clusters, until the number of clusters reaches the number of distinct points. Consider, after all, that if you were to place one cluster centroid at each unique point, then the distance of each point to its centroid would be 0, but that for any lower number of clusters, you could always improve the clustering by adding a centroid that matches exactly the location of one of the existing points that is not currently exactly at a centroid. Therefore, according to most definitions, the optimal number of clusters is the same as the number of distinct points.
