Hierarchical approaches, where nodes are clustered based on their network distances, have been shown to allow for robust and scalable topology-aware overlays. Moreover, recent research works have shown that cluster-based deployments of Internet Coordinates Systems (ICS), where nodes estimate both intra-cluster and inter-cluster distances, do mitigate the impact of Triangle Inequality Violations (TIVs) on the distance predictions, and hence offer more accurate internet latency estimations. To allow the construction of such useful clusters we propose a self-organized distributed clustering scheme. For better scalability and efficiency, our algorithm uses the coordinates of a subset of nodes, known by running an ICS system, as first approximations of node positions. We designed and evaluated two variants of this algorithm. The first one, based on some cooperation among nodes, aims at reducing the expected time to construct clusters. The second variant, where nodes are selfish, aims at reducing the induced communication overhead.