The results from an experimental evaluation study of cognitive network routing applied to the transmission of a scientific dataset and assuming a delay-tolerant space communications context are reported in this paper. The routing method that was tested relies on a spiking neural network (SNN) both to encode network performance predictions and to select the optimal outbound link for each data bundle. By defining an adaptive approach to bundle routing, the method can cope with changes in both the network congestion levels and the availability of redundant paths. Repeatable experiments were carried out using a laboratory network testbed to characterize the average routing performance. The reference implementation of Contact Graph Routing from the Interplanetary Overlay Network (ION) distribution was used to obtain the baseline performance of the current state of practice in space network routing. The results demonstrate the performance advantages of the cognitive networking approach to bundle routing when the network is being affected by different kinds of impairments, such as channel losses and link disruptions.