Flat DHT architectures have been the main focus of the research on DHT design so far. However, there have been also a number of works proposing hierarchical DHT organizations and pointing their advantages. They mostly rely on the intuitive understanding that hierarchy is desirable in any complex system. In this paper we formalize this intuition within a general cost-based framework. We provide a cost model of a specific hierarchical DHT organization composed of superpeers and leafnodes, and show that the costs of running the network are not necessarily minimized for flat DHT organization, providing thus a formal motivation for hierarchical DHTs. We further hint on what distributed algorithms can be applied in practice to reach optimal operating point of the network.