There is a class of data, including messages and business workflow state, for which conventional monolithic databases are less than ideal. Performance and scalability of Application Server systems can be dramatically increased by distributing such data across transactional filestores, each of which is bound to a server instance in a cluster. This paper describes a high-performance, transactional filestore that has been developed for the BEA WebLogic Application Server™ and benchmarks it against a database. The filestore uses a novel, platform-independent disk scheduling algorithm to minimize the latency of small, synchronous writes to disk.