Sep Kamvar has done a significant amount of research addressing precisely the problems you describe. The conclusion being that (in non-pathological cases, which would likely include the case you vaguely describe) one can design network architectures using reputation mechanisms such that malicious peers are found and routed around. See http://kamvar.org/publications, specifically "Numerical Algorithms for Personalized Search in Large-Scale Self-Organizing Information Networks" (the first Book) esp. Chapter 9, I highly recommend it.