Mobile information technology can help first responders assist patients more quickly, reliably, and safely, while focusing resources on those most in need. Yet the disaster setting complicates reliable networked computing. The WIISARD client-server architecture provides mobile IT support for medical response in disasters. Cached remote objects (CROs) are shared via publish/subscribe, enabling disconnected operation when out of network range and ensuring data consistency across clients with rollback/replay. CROs also provide a flexible, familiar, and performant programming model for client programmers. A drill with the San Diego MMST showed that a basic client-server architecture, even with CRO's, is insufficient, because prolonged network failures-to be expected in disaster reponse-inhibit group work. We describe an extension of the CRO model to clusters of computers that supports group work during network failures.