Micro-kernel architecture of NT incurs microbenchmark penalities---worse cache locality.
NetBSD does not have an efficient graphics platform.
Unix-centric results. Is this a fair set of applications?
Problems of translating microbenchmark behavior into explaining application behavior.