The short version:

"In TrustDavis, the ability to issue multiple identities or to change identity does not provide a significant advantage to a malicious party, since a malicious party must back each identity with funds that other players can use to protect their transactions".

"Individuals with no references can join TrustDavis through the use of security deposits."

"There should be some incentive for parties to provide references and take on risk. Thus, parties can function as insurers and charge a premium for the references they provide."

From that, you can see where this is going. The basic problem with online reputation systems, of course, is that creating a new pseudonym is usually quite cheap. This is a way to make it more expensive.

From a marketing perspective, it's not going to fly. If you had to put up a deposit to buy on eBay or to get a Gmail account or to post on Facebook or Hacker News, few would bother. There are, however, dating sites which require this of men.

There's also EigenTrust, a competing algorithmic approach: http://en.wikipedia.org/wiki/EigenTrust

i've been working for a while on this:

https://github.com/neyer/dewDrop

I've skimmed the paper. What if the refers default on their liability, too? Could a big enough net of cascading defaults cause unbounded damage?