The United States Department of Defense (DoD) employs virtual and game-based training simulators to train its servicemembers on both individual and collective cognitive and psychomotor tasks. The current employment of these training simulators is typically conducted in a stand-alone manner, with distributed simulation remaining the exception due to interoperability challenges. The OpenSimulator (OpenSim) is a popular open-sourced virtual world simulator that currently provides a persistent three-dimensional social community for its users. Under the Army Research Laboratory's (ARL) Military OpenSimulator Enterprise Strategy (MOSES) research program, OpenSim is being developed to serve as a prototypical distributed military virtual training environment for tactical operations. Virtual worlds for military training is an emerging domain. As such, detailed analysis of critical architecture parameters is required in order to optimize the performance of both the simulator's servers as well as the multitude of client connections. Unfortunately, due to a lack of extensive virtual world performance analysis, OpenSim server administrators often make arbitrary resource allocations to support their environments and training scenarios. Negative consequences to this approach are that typically too few resources are allocated to an overwhelmed server, resulting in an unresponsive environment, while too many resources are allocated to an underutilized server, when those resources could be more effectively applied elsewhere. In this paper, we analyze network bandwidth's effect on virtual world simulator performance so as to support the future creation of a predictive model that will determine the optimal amount of resources required to support a target number of concurrent users in the virtual world. This analysis, and the future development of our predictive model, will provide the OpenSim developer community with the knowledge required to best allocate resources to support expected server load.