Most classical stochastic inventory models assume that replenishment lead times are independent of the amount of orders placed. This assumption is clearly problematic in capacitated production systems, where queueing models show that mean cycle time is a nonlinear increasing function of resource utilization. We use a simulation environment created for rapid prototyping of supply chains to examine the problem of setting safety stocks in environments with workload-dependent lead times. We show that even when extensive historical data is used, conventional approaches to setting safety stocks fail to provide the desired service level when lead times are load-dependent, and suggest directions for future research.