I work on a podcast (only 7-odd episodes in, though I have a background in online/radio for an int'l broadcaster) which marries non/fiction with predominantly electronic musical composition and what with my lack of expertise passes for sound design. It's called Paper Radio [paperradio.net]. The sound is an important part of the storytelling, so my mixing priorities are fairly balanced between voice and 'sound'.
I use a combination of Waves Renaissance Vox and Compressor on voice and instrument submix busses respectively. I also sub-group certain source groups (at the moment I've separated SFX, raw interviews, studio voiceover, synth tracks, rhythm tracks, guitar and cassette tapes) and compress or limit them depending on the material. And I use L1 or L2 on the master channel.
But there aren't any particular ratios or thresholds I apply regularly. I change my settings every time according to what's there. The biggest challenge for podcasts, to my ears, is headphones or speakers.
It's fine if you're mixing voice program only, or voice with backing music — you can afford to lose the music a little. But if you want a clear and even balance of voice and music/sound, at roughly equal volume, I've found that's something difficult to achieve for listeners on both headphones (especially crappy iPod earbuds) and speakers.
Speakers tend to lose the music while headphones can make it so loud that it gets in the way of the voice.
I agree with @Bruce in that excess compression really can destroy the dynamics and texture of a piece; it can also make it extremely annoying to listen to. Unfortunately I have found that if I don't compress/limit material enough, people complain that our podcasts aren't properly audible on the peak hour train.
It's a bit of a conundrum and not one I can readily answer, although most of you would probably suggest the right level of dynamic compression will balance things out. My heart really favours a less compressed sound, but it doesn't seem to be easy to maintain peoples' attention with a big dynamic range, especially with similar podcasts being more of that overcompressed broadcast style.
Anyway, sorry for this rambling answer (I haven't slept much lately). In terms of file compression, I publish either 160kbps or 192kbps 44.1k/16-bit/stereo mp3s. As someone above pointed out, Flash can only stream certain sample rates (I think it's 11/22/44 - no 32/48k). The file sizes are a bit bigger but it's rich content that I think justifies the detail. If I were using voice only I'd definitely consider 128kbps and possibly even mono.
The only other thing I'd suggest is to consider who/what/where your audience are and what suits them. There isn't any real reason aside from slight added convenience why you wouldn't publish a great voice recording in 192kbps mp3 for a Scandinavian middle-class audience, but if you were podcasting primarily to an audience in the South Pacific, file size would be super important and you'd want to consider even mono 56kbps for some material, even with music. I believe in user-centred approaches for all situations where information must be communicated. A lot of work I've produced (both sound and otherwise) has been for a Pacific Islands audience, and the bandwidth and tech they can access has been instructive to say the least.
Again, apologies for my rant. I hope it helps in some way.
