In this paper, we extend and summarize our earlier work on optimal insensitive load balancing. Based on the insensitivity results by Bonald and Proutiere, we study insensitive load balancing in data networks executed either at packet or flow level. When insensitive load balancing is used, the flow size distribution does not affect the state distribution or performance of the system. The most efficient capacity allocation policy can be determined recursively when packet-level balancing is used. The flow-level balancing problem is analyzed using the theory of Markov decision processes. By using the linear programming formulation of MDP theory, the optimal routing policy can be solved either separately or jointly with the capacity allocation. Performance of the different methods is compared in a toy network. Packet-level balancing performs best of the insensitive policies. The performance of flow-level balancing is improved if capacity allocation and routing are jointly balanced and optimized. However, the requirement of insensitivity still implies some performance penalty in comparison with the best sensitive policy.