Waveforms transmitted by the elements of a MIMO radar system may differ in power and bandwidth. This raises the question of optimal resource allocation among the radar elements. Specifically, we are asking, given constrained resources, what is the optimal bandwidth and optimal joint power and bandwidth allocation for best target localization performance. Using the Cramér-Rao Lower Bound (CRLB) as the figure of merit for localization accuracy, the resource allocation optimization problem turns out to be non-convex. We apply a Difference of Convex functions programming approach to develop quasi-optimal algorithms for solving the resource allocation problems. A lower bound is also developed to help assess the quality of our solutions. Numerical examples demonstrate that bandwidth allocation has considerably more impact on performance than power allocation, and that the best performance is obtained with joint power and bandwidth allocation.