Linux has emerged as the system-of-choice in academic and production scientific computing settings. Much of the functionality in the Linux kernel, however, has been designed to arbitrate between competing applications thus ensuring safety and fairness. In scientific computing settings, particularly where resources are space-shared and "batch" scheduled, an executing application has exclusive access to the resources it is using. As, much of the Linux mechanism devoted to the arbitration of sharing is unneeded, and any performance degradation it introduces is purely perceived as overhead. In this paper, we investigate the performance effects of specializing the Linux kernel for the dedicated execution of scientific applications. We analyze the kernel execution behavior of a publicly available weather modeling application and propose two different specializations that are designed to improve its performance. We are able to achieve a maximum performance improvement of 24% over a "stock" Linux installation though our techniques, thus illustrating the potential power of this approach.