File transfers are and will in the future be responsible for a substantial part of the Internet traffic. However, with present solutions transfers of popular files lead to a lot of redundant data transfers in the network. In this paper, we investigate how a link level caching scheme can reduce the number of redundant data transfers. We serve requests from clients that download a file concurrently, but arrived at different times in such a way that they get at a given point in time the same data chunk of the file. This enables link caches to efficiently remove the redundancy. The data chunks are rearranged at the client to compose the original file. Through implementation and experimental studies we show that this approach clearly outperforms traditional file servers in terms of file server capacity and bandwidth consumption; especially when encoding the original file with fountain codes.