In this paper, moving target tracking performance in multiple input multiple output (MIMO) radar systems with distributed antennas and non-coherent processing is studied. Due to the use of multiple, widely distributed antennas, MIMO radar architectures support both centralized and decentralized tracking techniques. Each receiving radar may contribute to central processing by providing either raw data or partially/fully processed data. Estimation performance of centralized and decentralized tracking is analyzed through the Bayesian Cramer-Rao bound (BCRB). The BCRB offers insight into the effect of the radars geometric layout, the target location, and propagation path losses on tracking accuracies. It is shown that, with different propagation path loss, the manner in which decentralized estimations are combined in the fusion center effects the overall estimation performance. Two tracking algorithms are proposed, corresponding to respectively, a centralized and decentralized modes of operation. It is demonstrated that communication requirements and processing load may be reduced at a relatively low performance cost. Based on mission needs, the system may use either approach: centralized for high accuracy or decentralized for resource-aware tracking.