A number of agents asking for resources distributed over networks can be modelled as a n-party game. A resource broker defines the game payoff matrix. Linearly transformed agents' payoffs define their utilities. Taking statistical ensemble of such systems, with agents playing the role of subsystems, one can treat it in the framework of statistical thermodynamics, where an overall system's utility is an analog of the energy function. However, for games with no predefined agent resources, the systems are correlated, the probability measure is not Boltzmann-like, and the generalized, nonextensive thermodynamics has to be incorporated. Among many fields of application of this scheme, job scheduling on a computing Grid with distributed resources is found particularly interesting.