Actually i see that quite a bit. All I ever use is Quats so i just figured that's part of using Quats.
The more distantly a bone is rotated from its original position and is rotated on more than one axis, the less the curves seem to correspond to the color of the manipulator rings. I don't know why, but they don't.
I stopped being bothered by it once I realized that the actual value of a red curve or a blue curve isn't something i needed to worry about anyway. What I really needed to see was whether the curves were flat when I needed them flat and whether they were steep when i needed them steep. That's easy enough to edit even if the green curve is doing it instead of the red.
If I need the bone to be in a certain angle at a certain frame I just move it to where I want it to be and the curves move to where they need to be on their own.
It's a bit odd that a toe bone (?) which hardly gets rotated very much at all is doing this but but I dont' know where that toe has been!
Just to experiment, select that one bone, turn off all but the rotate filter, press the "compensate mode" button and then the keyframe button to send the bone back to what it believes is its default rotation and see what that looks like. That should zero all the curve values out.