As XML has become the de-facto standard for data exchange in several applications, XML message brokers assume a lot of importance. Message brokers we address in this paper store a large number of user profiles, in the form of XML twig queries, that represent the data requirement of users/applications. In contrast to publisher-subscriber systems, message brokers select the part of the data that is of interest to the user and send it to her, instead of the whole matched document. In this paper, we propose a new sequencing based XML message broker architecture that handles tens of thousands of user profiles and delivers to each user the XML elements (or nodes) that satisfy the user's twig query. We introduce a novel node sequencing method called branch sequencing that converts an XML twig query into a branch sequence. Intuitively, nodes are sequenced branch by branch. We show that by storing user profiles in the form of branch sequences, and appropriately indexing them, we can efficiently generate the qualifying nodes for each profile, as the input XML document streams by. The input document nodes are streamed into the system in the document order. The branch sequencing technique we propose enables holistic matching of twig queries to the document. Another important feature of the system is that it carries out ordered profile matching. We have implemented the system and compared its performance with a state-of-art XML filtering system and have shown that the filtering time reduces significantly and the system scales up very well.