Bitstream Structure Descriptions (BSDs) allow taking the complexity of transforming scalable bitstreams from the compressed domain to the semantic domain. These descriptions are an essential part of an XML-driven video adaptation framework. The performance of a BSD transformation engine is very important in such an architecture. This paper evaluates the efficiency of XML-based transformation languages in our video adaptation framework. XSLT, STX, and a hybrid solution are compared to each other in terms of execution times, memory consumption, and user-friendliness. Our experiments show that STX is the preferred solution when speed and low-memory are important. The hybrid solution is competitive in terms of memory consumption and is more user-friendly than STX. Although XSLT is relative fast, its memory consumption is very high.