The recent trend of constructing application-level overlays makes it increasingly common to have end-to-end paths decomposed into multiple overlay segments (tunnels). This paper investigates the potential benefits of making these intermediate overlay nodes both application- and congestion-aware in the context of streaming media. In particular, we demonstrate that hop-by-hop congestion control improves the quality of TCP-friendly delivery of layered video applications by up to a factor of four in the steady state. Moreover, the resulting system is significantly more responsive to changes in capacity, all the while not negatively impacting competing TCP flows.