We investigate thread-level concurrency in several common desktop applications. We find that the majority of active periods (periods of uninterrupted CPU activity for a single thread) are relatively short, while the few long active periods account for most of the active time. The shortest 90% of the active periods only account for roughly 12.75% of total active time. We speculate that this is generally true for most applications, and that there might be some way to take advantage of this fact in the scheduler. Due to the difficulties in catching, testing, and fixing concurrency bugs, we propose modifying the thread scheduler to reduce the risk of concurrency bugs where possible. Our simulations show that our modification may work well for certain applications depending on the level of CPU demand.