Before continuing, I do want to say I am +1 on the bounty system in principle. However the location bothers me.
I would like to comment further on the advantages and problems of the following options (I am not saying that the problems cannot be addressed):
1- attach to current wishlist tracker (bounties as direct extension of wishlist)
Main advantage: visibility, integrated, can see bounty info at one glance.
Possible problem 1: Some feature requests are very important from a code perspective, but not particularly attractive from a bounty perspective. The important stuff is left undone or delayed.
Possible problem 2: Some feature requests are very important to the community (as reflected by the voting), but receive low bounty. Others are needed only by someone who happen to be rich (hence high bounty but low voting). This could get awkward.
Possible problem 3: I think bugs should not be bountied. Even pledge of $20 to solve bug might cause crowding out of intrinsic motivation to fix bugs. This is my gut feel and also supported by economic theory that predicts crowding out. See http://www.slideshare.net/nice/crowding-effects-how-money-influences-open-source-projects-and-its-contributors/ By having all the bugs/feature requests and bounties together, the likelihood of crowding-out of intrinsic motivation could be very high.
2- bounties on separate site (no information on bounties on dev)
Advantage 1: The community has a stronger moral obligation to fix or add those requests that are not high bountied but still important. If you have the bounty information too close (using option 1), even if you don't say it, it might imply some kind of moral statement that bounty value is a good gauge of importance, which is not in many cases.
Advantage 2: Developers who work on important but low-bountied items are perceived as more highly valued if bounty information is not on dev. Otherwise, it may look to the less enlightened viewer that less-well-bountied == less important which is totally not true.
Advantage 3: There may be developers who prefer not seeing what others are earning all the time. It may be a distraction from their intrinsic motivation.
Possible problem 1: Poor visibility leading to developers not knowing that there is in fact a bounty for an item. The good case would be that the dev does something important. The bad case is the nothing else is done.
Possible problem 2: Too many sites, causing confusion.
Possible problem 3: More work, more cost.
3- bounties on separate tracker in dev (this is a compromise between the above 2 options)
This helps to address the problems in option 1, but IMO, still insufficient. The "moral values" element concerns me deeply. I'm afraid we will give the impression that cash determines what's important, which is not true.
I look forward to more comments and suggestions.