XML data sources are more and more gaining popularity in the context of a wide family of Business Intelligence (BI) and On-Line Analytical Processing (OLAP) applications, due to the amenities of XML in representing and managing semi-structured and complex multidimensional data. As a consequence, many XML data warehouse models have been proposed during past years in order to handle hetero-geneity and complexity of multidimensional data in a way traditional relational data warehouse approaches fail to achieve. However, XML-native database systems currently suffer from limited performance, both in terms of volumes of manageable data and query response time. Therefore , recent research efforts are focusing the attention on fragmentation techniques, which are able to overcome the limitations above. Derived horizontal fragmentation is already used in relational data warehouses, and can definitely be adapted to the XML context. However, classical fragmentation algorithms are not suitable to control the number of originated fragments, which instead plays a critical role in data warehouses, and, with more emphasis, distributed data warehouse architectures. Inspired by this research challenge, in this paper we propose the use of K-means clustering algorithm for effectively and efficiently supporting the fragmentation of very large XML data warehouses, and, at the same time, completely controlling and determining the number of originated fragments via adequately setting the parameter K. We complete our analytical contribution by means of a comprehensive experimental assessment where we compare the efficiency of our proposed XML data warehouse fragmentation technique against those of classical derived horizontal fragmentation algorithms adapted to XML data warehouses.