hairy_hen said:

RMS stands for "root mean square". Rather than a simple average, it involves finding the average of the squares of the values being measured, then taking the square root of this value. I believe Dolby recommends using this method for Dialnorm because it more accurately reflects the characteristics of human hearing.

A RMS measurement does not take into account "perceived" loudness. Dolby Labs recommend that a long-term A-weighted average of dialogue is used as a loudness measure for dialogue normalization.

The problem is of course, that the more conservatively mixed tracks have a low dynamic range when compared to the 70mm/'93 mix. If all files have waveform peaks at digital 0dB, then the average level of the the mono mix, and the '77 and '85 Dolby Stereo mixes will be much higher than that of the 70mm/'93 mix.

You can read about the fun and games I had setting dialnorm for the Pwnage DVD here. Even if you set dialnorm by the book for dialogue, music ends up being the dominant factor in requiring the volume to be turned down when switching audio tracks.