I don't think you're going to get any feedback/echo with USB headsets. And honestly, I am all for dong a great job on whatever I'm working on, but I just feel that you can get very similar results by just recording the convo as a whole on one end vs delivering separate recordings, assembling said recordings, treating each host with EQ, etc, with a shitload less work. Also, if someone happened to one of the host's recording process during the podcast, that would fuck everything up. I feel all that work could be spend editing the thing which takes a lot fucking longer for podcasts than you would imagine.
I think it's really up to the people hosting the podcast and if they want to set it up to record their mic and deliver it. If everyone is down, and who ever would be syncing/editing is down then I don't see why it is a problem.
I'm not trying to shit on your idea, but just trying to say that the benefits you would have in post by getting everyone's audio separately wouldn't outweigh the effort required compared to just recording the convo from Skype/Vent.
We could just run tests to see how everything sounds and try out different recording methods.