Multicomputers have traditionally been viewed as powerful compute engines. It is from this perspective that they have been applied to various problems in order to achieve significant performance gains. There are many applications for which this compute intensive approach is only a partial solution. CAD, virtual reality, simulation, document management and analysis all require timely access to large amounts of data. This thesis investigates the use of the object store paradigm to harness the large distributed memories found on multicomputers. The design, implementation, and evaluation of a distributed object server on the Fujitsu AP1000 is described. The performance of the distributed object server under example applications, mainly physical simulation problems, is used to evaluate solutions to the problems of client space recovery, object migration, and coherence maintenance. The distributed object server follows the client-server model, allows object replication, and uses binary semaphores as a concurrency control measure. Instrumentation of the server under these applications supports several conclusions: client space recovery should be dynamically controlled by the application, predictively prefetching object replicas yields benefits in restricted circumstances, object migration by storage unit (segment) is not generally suitable where there are many objects per storage unit, and binary semaphores are an expensive concurrency control measure in this environment.