Technological developments of the last few years have favoured the creation of distributed networking infrastructures (usually referred to as community networks) where the resources are made available to the members of a community of people. With emerging large-scale community infrastructures the opportunities for new commercial services and for innovative business models are becoming feasible. Since one of the most resource-demanding services today is access to user-generated content through the Web, suitable content delivery services are needed in the context of community networks in orders to make an effective usage of shared resources. In this paper we describe two architectures we have designed to provide optimized delivery of multimedia Web content (e.g. video but also other kinds of user generated content) within community networks. Both architectures work without any kind of cooperation from the original content providers but assume that a basic Web caching service is provided within the network, either by commercial service providers or by the community members themselves. While the first architecture relies on a traditional centralized control entity, the second is designed according to the peer-to-peer paradigm, in order to provide better scalability, robustness to failures and self-configurability. Experimental results are also presented aimed at evaluating the performance gains for end-users from a localized distribution of content in scenarios in which community members are distributed in clusters sparse at different geographic locations.