Recently I faced a need of slightly more efficient approach to compute clusters from a point cloud.
Cluster SOP seems to be a great tool, but unfortunately is a bit slow, and it doesn't take the advantage of all cpu cores.
Here you can find an attempt to make cluster calculation a bit faster. Code was used for different purposes, but to make your life more easier I wrapped it as a SOP operator.
k-means++ is not implemented yet, I might consider implementing it in the future.
My research was based on Peter Goldsborough's tutorial.