Tonight I decided to do a little controlled experiment with mix compression.
I made sure tonight to really play around with different compression ratios and thresholds to see if I could notice any difference in "loudness". I tried some ratios as low as 1.5 and as high as 5.5 and tweaked the attack/release to something that sounded proper (usually ended up that the attack was around 10ms and release around 150ms). Gain reduction usually remained steady at around 3 (per andrew's advice) and peaked at around 7. I also turned the mix comp completely off and tried to see where I could get my "loudness" with peak limiting only.
My ultimate conclusion is that after applying some reasonable peak limiting, I wasn't able to notice any difference in "loudness" between the mix comp being on or off. When it came down to it, I was applying so much individual compression at the track level that the only thing the mix comp did, in my opinion, was add muddiness. It's a bit embarrassing that I got so caught up with the preconception that mix bus compression equals loudness that I never bothered to shut the damn thing off during my mix. You live you learn!
As promised earlier, here is a copy of the song without any mix bus compression. This is likely the one we'll go with.
http://dl.dropbox.com/u/14404243/4Unsur ... nocomp.mp3