Delay tolerant networking (DTN) exploits the mobility of nodes to serve as message ferries for content delivery. It is the goal of message ferry selection to choose a node that soon moves close to the destination. Proper ferry selection can strongly affect system performance. We analyse this impact for data intensive applications in emergency and rescue (ER) operations. Through extensive simulation studies capturing the entire protocol stack, we compare approaches that have a priori static knowledge about the typical mobility patterns in such operations with evolving knowledge based approaches. Our results show that evolving strategies achieve almost the performance of approaches with static a priory knowledge in ER scenarios at the cost of control overhead, especially in non-DTN like dense networks. The static approaches are prone to malfunction if real mobility strongly deviates from the assumed mobility pattern. We demonstrate that it is important to simulate the entire protocol stack for such applications in DTNs. Delay tolerant forwarding is strongly affected by the underlying layers. Simulation results would be of low relevance for real-world implementations without modelling them.