The packet-pair structure is a widely used probing technique for measuring link capacity over network paths. In packet-pair probing, the probe-gap value at the link of interest (e.g., the link with the smallest capacity or any link along a path) is determined for link-capacity measurement, which can be affected if dispersion (i.e., separation) of the packet pair is created by the heterogeneous link capacities and cross traffic over an end-to-end path. In this paper, we present an analytical model concerning the effect of heterogeneous link capacities and cross traffic on the packet-pair structure over a multiple-hop path, and define the constraints on the packet sizing for accurate end-link (i.e., the link connected to the destination node of an end-to-end path) capacity measurement. We test the model through numerical evaluations and experimentally, on end-to-end paths with heterogeneous links and different cross traffic loads. The experiments are performed on a testbed and in the Internet.