Unless I have broadcast specs that say otherwise I try to have my peaks around -6Db. This of course is just my guideline. In reality you can take it as close to 0Db as you want for web and live event playback, I just like to leave a bit of headroom.
As for mixing music, that is a job for your ears. There are no hard and fast rules that I know of when it comes to setting the level. My guideline is that I always go a little lower than I think it needs to be. I base this on the fact that I know exactly what the dialogue is saying from editing it so it's easy for me to pick it out.
The only other trick I can give you for music is a channel EQ. Apply the Channel EQ to a piece of music and then set a couple keyframes so that you can adjust the ranges between 200Hz and 4000Hz. These are the general frequencies where the human voice resides. Then bring that whole rage of frequencies down about 3 to 5Db. This basically reduces the level of the frequencies in the music that are in the same range and the voice. By doing this you can then bump the music up a little more so it doesn't loose it's impact.
I think for the web most people like to let the peaks go all the way to 0. I'd suggest using a compressor to make a more consistent volume. I generally try to keep the music around -20db when there's dialogue.
OutOfFocus.TV - Original series, music videos, and entertainment for your couch.