Volunteer Grids represent the most dynamic and unpredictable computing platform in the Grid space. Nodes may come and go, offer varying capabilities, and varying levels of reliability. Despite these "features" this is an attractive platform for scientific computing due to its scale and ease of management. Our research aim is to harness these open systems for a broader class of high performance e-science applications. We discuss one of the main obstacles to achieving this goal - how to effectively manage the inherent unreliability in the system.