Nodes in a network often store and serve content to other nodes. However, each has a finite capacity and if requests for content exceed the capacity then queries fail. It is generally not possible for a priori predictions of load demand because at given times some content may suddenly become popular and at others hardly requested at all. Hence over a given time period a population of nodes has a certain total capacity to serve requests (the sum of all individual node capacities) and some demand load (queries going to the nodes). Assuming nodes can replicate content and redirect queries we present a simple node level protocol that self-organises nodes into cooperative clusters leading to efficient outcomes in some simple scenarios. The approach is scalable, robust and self-organising but there are several open issues. We present this as work-in-progress.