We have been working hard running benchmarks on multiple platforms, including machines we have in our lab as well as on systems provided for our testing use by HP, (which we greatly appreciate). As we discuss our findings, we have been struck by how often we hear that we should run multiple instances of the JVM to address the issues we have found to work on. While this is good advice for those running app servers, for example, it does not address our approach, where we want to apply as much compute power as possible to the single task we have parallelized. One of our engineers found this article outlining the performance characteristics of multiple vs. single virtual machines, which helped our understanding. "When using SPECjbb2005 to compare software and hardware configurations, its critical to run and compare both single-JVM and multi-JVM configurations, and from this point forward I plan to advocate that Sun continue as it has done today and submit both single-JVM and multi-JVM configurations on new hardware platforms. "