We believe that improving an operating system's support for the evolution of software is vital to our goal of reducing the significant sum spent on adapting existing software to changing user requirements, especially to improve the performance of software. Therefore, we proposed the idea that by increasing an operating system's abilities to observe the software's execution behavior and evolve its execution behavior using observed results, an operating system could adapt existing software to changing user requirements without making any changes to the software. We integrated the above abilities into a CPU scheduling mechanism in an operating system, and verified the usefulness of our idea using existing software, i.e., a World Wide Web (WWW) server. In this case, our scheduling mechanism alters the execution behavior of a WWW server by giving preferential use of the CPU resource to server processes handling HTML file requests. This allows the user requirement, which is the enhancement of response time during periods of high demand, to be satisfied. In order to determine which processes are server processes handling HTML file requests, we introduced scheduling parameters SLP and RW. In this paper, we describe how we predicted and updated parameter RW based on the observed execution behavior of a WWW server, and present the experimental validation of our method.