The work a computer system performs to temporarily suspend execution of one process and resume execution of another process is called a context switch. The time it takes for a computer system to perform a context switch is often cited as an important performance metric. Lengthy context switch time can result in poor performance in process-intensive applications such as distributed servers or interactive systems. However, context switch time is usually assumed to be constant for a particular system. This document describes a benchmark program which shows that for many systems this is not the case. The program was run on a variety of UNIX��-based systems at NCAR, and the results are given in both tabular and graphical form. They show that for many systems, context switch time increases, sometimes dramatically, with the number of concurrent processes running in the system. This non-scalability of context switch times leads to diminishing returns when, for example, upgrading to a faster processor in order to support more interactive users.