Validating current and next generation of distributed systems targeting large-scale infrastructures is a complex task. Several methodologies are possible. However, experimental evaluations on real testbeds are unavoidable in the life-cycle of a distributed middleware prototype. In particular, performing such real experiments in a rigorous way requires to benchmark developed prototypes at larger and larger scales. Fulfilling this requirement is an open challenge but is mandatory in order to fully observe and understand the behaviors of distributed systems. In this work, we present our Virtualization environment for large-scale Distributed Systems, called V-DS, as an answer for reaching scales steps further than current experimental evaluation scales. Then, through the use of V-DS, we experimentally benchmark the scalability of three P2P libraries by reaching scales commonly used for evaluations through simulations. Notably, our results on the Grid'5000 research testbed show that only one of the P2P library can optimally build overlays, whatever the size of the overlay is