MP3 Conversion Issue: Inconsistent dBFS Reduction for Long Audio Files in Audition

And having thought about it some more, and recalling some research I did nearly 30 years ago, I can't say that I'm surprised that there are variations - it's in the nature of the beast. MP3 coding has two main methods of data reduction, although we generally only talk about one of them - the coding gain resulting from critical band frequency 'masking'. That doesn't make much of an alteration to coding levels, but the other mechanism does (that was the gist of the research). It's like this:

The coding algorithm also uses discrete cosine transforms to look at how repetitive the signal to be coded is, and if it finds repetition, it will just say 'well this bit is the same as the bit we just coded, so let's send a 'repeat' code instead coding it over and over again. This is a simplified explanation, but MP3 coding is seriously complicated to explain in detail, and is somewhat inappropriate, so I'll stick with that. The bottom line is that the effect of all this coding gain is that you will experience small changes in level, especially with continuous signals. The degree will vary according to the coder you use, and whether you used CBR or VBR, and that one's easy to answer - always use CBR, especially if you want the timing of your file to be anything like accurate. The reason that the coder makes a difference is that in the standard, it says that you can use any particular coder algorithms you like, but that all decoders have to be identical.