I've always used k-means and set k=n when I didn't know k ahead of time (+some center merging criteria, which seems to come down to h in eq. 1), but it's pretty expensive. From a cursory glance it looks like this is O(n) to my O(n^2). That said, k-means is a lot simpler--I imagine that's why. There's reasonably complex math (at least, non-intuitive), and you have to choose h. Everyone learns bubble sort, even though nobody uses it in practice (although given, everyone learns quicksort & mergesort too).