The proliferation of mobile devices and the fact that high-bandwidth and continuous connectivity is not available everywhere, has led to the creation of hoarding algorithms that attempt to mitigate the problems related with disconnected operation and with the operation in areas where bandwidth is either scarce or expensive. In this paper, we present a hoarding scheme for location-based data in semi-structured information spaces, such as the World Wide Web, which relies on clustering of semantically related data items. We show by means of experimental evaluation that our clustering-based approach outperforms existing hoarding techniques that do not make use of clustering by a factor of more than 2 in terms of hoard cache hit ratio.