Typical wireless routing protocols like AODV/DSR are not scalable to very large networks because they employ flooding for route discovery. Geographic routing protocols like GPSR are highly scalable because they require minimum control overhead, but depend on idealized link quality models (such as the unit disk model) which are not always applicable. We explore the routing spectrum between these two extremes under a realistic random link quality model. It is common wisdom that by adding limited flooding to a protocol like geographic routing improves quality. In this paper, we provide a formal and quantitative formulation of this trade-off, and show both analytically and experimentally that a significant improvement in path quality is possible by searching a narrow region around the geographic straight-line path between the source and destination. In particular, if the endto-end throughput is measured as the product of link reliabilities in a path, then we demonstrate that the path quality improves exponentially as the search region is broadened.