Some anecdotal evidence demonstrates success of the extreme programming practice in a portion of the software industry. It has also been argued that pair programming, as a part of the extreme programming process, yields higher quality software products in less time. On the other hand, these principles are sometimes questioned with respect to resource allocation and management issues. Although precise information about benefits and costs of the extreme programming practice represents a critical guideline for improvement of software quality, there has been little work on the subject beyond subjective reports and a study in an academic environment. We propose an experimental framework to quantify benefits and costs of the pair programming practice and compare design aspects of the resulting software products and their defect behavior. For this purpose, we use a set of object-oriented metrics and software reliability growth models based on service requests.