As I understand it, R128 uses a lower volume target than the previous method. Lower than a lot of people would expect, myself included. This keeps the volume constant across the the entire range of music available.
Adaptive Volume then looks at what you have playing and says, "how much louder can I make this, and not clip anything that's currently playing?". Then it adds that fixed amount, to bring back some of the gain. This number will change as you add louder music, or remove louder music. This is why it's optional: Because the absolute volume target, when using Adaptive Volume, changes from playlist to playlist.
It's a complex situation that took me some research to understand. I'm not even 100% sure I've got it right. But I think I do. Someone correct me if I'm wrong.
Brian.