Desktop Grid is an important part of the high-performance computing landscape. Basing on a Desktop Grid computing approach one could be able to gather huge computing resources to solve “bag-of-tasks” problems. However, the computing power of the server of a Desktop Grid, as well as some computing projects settings, could become the limiting factors to utilize the collected CPU cycles. In this paper we consider a Desktop Grid as a queueing system and characterize its performance; we introduce the concept of social costs and estimate the peak and real performance; we also highlight some aspects of effective usage of the computing power gathered under a Desktop Grid.