The problem with using bounties as transferrable currencies is that then it stops being a good measure of the reliability of an author--- the author that places bounties on everything seems more unreliable. It is possible to institute an "ask to answer" system where the reputation of the answerer is increased, but the reputation of the asker is not decreased. If you make the rep gain proportional to the rep of the person asking, and perhaps award points to both the asker and the responder according to the rating of the question/answer, you can encourage people to ask folks to answer questions, without harming the reliability metric aspects of reputation.
But I think that it is not really necessary, as bounties didn't really work to make the questions get an answer reliably, although sometimes it helped, it wasn't reliable, mostly because of the week time limit. Bounties for research level problems really should last indefinitely, perhaps even slowly increase with time.
You can replace bounties with real, actual, honest to goodness money. Erdos's created this system of "money problems" in his field, and these were a huge boon for mathematics. He would place a money bounty on problems, from 100dlr to 1,000dlr, and he would pay it out of pocket if someone solved it! He wasn't particularly wealthy either, it was an actual financial cost to him. But if you published a solution, you would know that a check would come in the mail from Erdos. This was a tremendous thing for combinatorics, a lot of problems got solved this way, and Erdos was very good at placing appropriate bounties according to the difficulty. People solved it, not because they needed the money, but because it was considered a high honor to have gotten a check from Erdos.
In this case, the Erdos money problems were all research level problems, although at various levels of difficulty. A 1000dlr Erdos problem was usually very difficult, requiring a major idea, while a 100dlr Erdos problem was approachable, but still difficult. I think that this model is very effective, and this way, 30 people who care about a problem can each contribute 10dlr, and make a 300dlr Erdos problem, and then the solver gets 150dlr, and this site gets 150dlr (but only if the answer is satisfactory and complete, of course).
This is really ok, if the site is a non-profit, and the money sums involved don't become astronomical, like the millenium prizes.