I am also working on a project to try to take advantage of Grid computing environments, perhaps better named "Internet computing environments," for large-scale scientific computation. The extreme processor and heterogeneites and hierarchies present new challenges, and make architecture-aware programming essential to achieve acceptable performance. I am part of the Rensselaer Grid group which is investigating these issues. I have worked with Carlos Varela and his student Kaoutar El Maghraoui at RPI on the development of IOS/MPI, which is an extension that allows process migration within an MPI program. Both DRUM and Zoltan's hierarchical balancing implementation will play important roles in extending large-scale parallel adaptive computation into Grid environments.
I have recently become interested in high-performance computing for bioinformatics, and I am part of the BiGP (Bioinformatics, Genomics, and Proteomics) program that has been formed at Williams.
Here's a picture from a research visit to Sandia, taken April 11, 2002:

Front row from left to right: Karen Devine, Pavel Bochev, Rich Lehoucq, David Ropp (all Sandians). Back row from left to right: Curtis Ober, Simon Tavener, Donald Estep, Jim Teresco, Max Gunzburger, Joseph Flaherty, Martin Berggren, Bill Hart, John Shadid, Roscoe Bartlett. (Ober, Berggren, Hart, Shadid, Bartlett-Sandians; Tavener and Estep, Colorado State University; Teresco, Williams College; Flaherty, Rensselaer Polytechnic Institute; Gunzburger, Iowa State University).