Distributing power packets in a network is an efficient technique for supplying power to distributed loads. In the case that the amount of power is limited in a closed network system, the consensus algorithm can be useful for sharing the limited power resource in the system. For example, we consider a power network consisting of one power source and N(≫1) loads. We assume that each load has a buffer that can store the power for driving the load. It is important to efficiently share the limited power from the single source. One possible way for sharing power is as follows. First, all loads are divided into several clusters. Next, the power source sends power packets to each cluster by responding to the feedback information on how much power is required by the loads in the cluster. If the power demand of one load exceeds the power supply from the source, the deficient power is accommodated by the consensus algorithm among the buffers within the cluster. We discuss by numerical simulations how to cluster the loads in terms of the balance between the cost for distributing power from the source and the performance of loads.