The use of inverse-multiplexing with diverse-path routing techniques in transport networks is expected to satisfy the traffic demands at the expense of lower capacity requirements. However, when the traffic between two nodes is routed over multiple paths, the different flows have different arrival times at the destination node. In order to compensate for this difference, known as differential delay, data has to be temporarily buffered in this node. Eventually, the size of the high-speed buffers involved may expand beyond acceptable figures, resulting in a cost increase that may even cancel the advantages attained with these techniques, in terms of network capacity. As an alternative to compensating for the delay only at the end-node, the use of distributed schemes can enable the reduction of the maximum buffer size per node, while preserving the capacity savings from diverse-path routing. Hence, this paper presents a novel model for optimizing the differential delay compensation across the network, when inverse-multiplexed diverse-path routing is employed. In order to quantify its gains, the framework is applied in the routing of 100 Gb/s Ethernet signals over high-speed optical transport networks (OTN) using virtually concatenated 40 Gb/s optical channel data units (ODUs). In the two reference network topologies evaluated, the buffer reduction obtained varied from 67% to more than 90%.