My guess is that it's adaptive volume that is the problem, you should try it with just volume levelling enabled.
Volume levelling normally works by attenuating tracks so that they're all around the same level. Adaptive volume in peak normalize mode adds back in as much volume as can be added without driving any track in the playlist into clipping. So if you were playing a mixed playlist of three files with peaks of 0dBFS and volume levelling targets of -10dB, -15dB, and -20dB, volume levelling would subtract those amounts from each file and adaptive volume would add back in 10dB or so because that's the most that can be added to every file without driving any file into clipping. Add a fourth file with 0dB peak and a -5dB volume levelling adjustment to the playlist, and adaptive volume will only add 5dB to the new playlist (because it's adaptive). It works a little differently with albums than mixed playlsits, but the basic operation of adaptive volume is similar.
The bottom line is that when used as DSP options at runtime, adaptive volume and volume levelling work together dynamically on a playlist, but they will work a little differently on any given playlist. I have no idea how the logic works during a sync because there's no "playlist" to use as a reference for adaptive volume. I'm guessing that adaptive volume just looks at each track individually and adds back in the boost that volume levelling removes which effectively breaks volume levelling in mixed playlists entirely; If you want consistent volume levels in the synced files, try just using volume levelling and see if that works better (it has always worked fine for me).