Recent years have seen a trend towards unmanned multi-sensor surveillance networks with large numbers of cheap and limited-performance sensors. While these networks hold significant potential for surveillance, it is of interest to address fundamental limitations in large-scale implementations. We first introduce a simple analytical tracker performance model. Analysis of this model suggests that scan-based tracking performance improves with increasing numbers of sensors, but only to a certain point beyond which degradation is observed. Correspondingly, we address model-based optimization of the local sensor detection threshold and the number of sensors. Next, we propose a two-stage tracking approach (fuse-before-track) as a possible approach to overcoming the difficulties in large-sensor surveillance, and we illustrate promising performance results with simulated surveillance data.