When you are mixing dialogue and you have it all mixed, what is your strategy for compressing the rest of it? I understand that multiband compression is commonly used but do you try to set a very high ratio and bring up the human vocal frequency up a few decibels, or keep a low ratio and allow the end chain limiter to take care of peaks. I'm mixing my Dialogue for a project at -24dbfs. I have some peaks in these tracks that get rather high and at some times throughout the mix there are ratios of 10 and make up gains that are sometimes 13 or more. I've used multiband compression on many of the tracks for noise reduction and downwardly expanded most tracks to cut off some noise floor but still retain the full sentences of dialogue. It seems like maybe if you are combining these compression ratios, it would reach the point of limiting but I also don't know if it works like that.
-24dbfs seems very low for peaks. Average maybe, or LKFS, but low for peaks. What spec are you trying to hit?
How much I compress the dialog depends on the deliverable. For films/DVD then I try to get away with as little as possible - none if the style/production audio allows it. In those cases I'll go more for "character" (i.e changing the sound of the dialog) then level control. For TV Broadcast, well, you're just not going to stay in QC range without using it - at least under the time constraints that TV usually has.
As you mentioned, I'll try and use several stages of compression - each doing no more than 2 or 3db of reduction - for example I might use an automated level control (i.e WaveRider, Vocal Rider) on the track - "character" compression ( wt slow attack and mid release) to help "even out" the levels on the bus, maybe a little multiband to help smooth out ragged frequency response (esp. with prod. dialog that's all over the map) and then a limiter to insure I'm not going over the QC levels. That would be a worst case scenario - but, when you've got 4 hours to mix a half-hour show, you do what needs to be done.