Multi-layer ultra-dense satellite networks (MLUDSNs) have soared this meteoric to provide vast throughputd for globally diverse services. Differing from traditional monolayer constellations, MLUDSNs emphasize the spatial integration among layers, and its throughput may not be simply the sum of throughput of each layer. The hop-count of cross-layer communication paths can be reduced by deploying inter-layer connections (ILCs), augmenting MLUDSN's throughput. Therefore, it remains an open issue how to deploy ILCs to optimize the dynamic MLUDSN topology to dramatically raise throughput gains under multi-layer collaboration. This paper designs an ILC deployment strategy to enhance throughput by revealing the impacts of ILC distribution on reducing hop-count. Since deploying ILCs burdens the satellite with extra communication resource consumption, we model the ILC deployment problem as minimizing the average hop with limited ILCs, to maximize throughput. The proposed problem is a typical integer linear programming (ILP) problem, of which computational complexity is exponential as the satellite scale expands and the time evolves. Based on the symmetrical topology of each layer, we propose a two-phase deployment scheme to halve the problem scale and prioritize stable ILCs to reduce handover-count, which decreases the exponential complexity to a polynomial one, with 1% estimation error: Simulation results based on realistic megaconstellation information confirm that the optimal number of ILCs is less than P.S/2, where P and S are orbits and satellites per orbit. Besides, these ILCs deploy uniformly in each layer, which raises over 1.55x throughput than isolated layers.