Traditional reliable transport protocols have been designed to perform end-to-end delivery transparently to the intermediate nodes along the path (e.g., TCP). The resourceconstrained nature of a Wireless Sensor Network (WSN) requires a new paradigm where intermediate nodes are able to cache packets and retransmit them on-demand during packet loss recovery in order to avoid incurring costly end-to-end retransmissions. This paper addresses the problem of cache partitioning optimization at the transport layer. We implemented several caching policies integrated into the DTSN transport protocol and analyzed their effect on overall network performance in terms transmission cost, throughput, and fairness. Our simulation results show that the choice of cache partitioning policy has a very significant impact on the performance and motivate the need to develop more advanced policies that adapt to dynamic network conditions and more complex topologies.