An interesting article, yet the research was done only on Gnutella. So what's new here?
The idea of "peer review" was only briefly mentioned, but it seems to me that dynamic, community-controlled criteria of trust are the best way to drastically prevent malicious damage. There can never be absolute trust, as was discussed in the article, but giving ordinary peers the power to define their optimal environment would both compartmentalise and restrict any peers or bots with malicious intent.
There should be some kind of democratic "voting" system in place, but not to the detriment of unusual requests (which often deal with rare content.) For the common items on Gnutella it may be fine, but IMO there has to be a better solution than that!
Thanks for the article anyway.