This explains it better than I can:
That's an interesting explanation!
I think there are two things at work here. One factor is the apparent shortening of the sway bar ends (for example, L1 vs. L2 in your diagram). If you look at adjustable sway bars, one of the most popular techniques is simply to allow the attachment point of the link (to the sway bar end) to be adjustable (they can be adjusted back and forth on the ends, essentially shortening or lengthening the "lever arm"). If they are adjusted in a direction that lengthens the distance between the link and the bar, the sway bar becomes soft. If they are adjusted closer to the bar, the bar is stiffened (because of the shorter lever arm). In effect, this is the same thing that happens when the bar runs off horizontal. The "effective" length of the "end" is shortened and hence the "effective" lever arm is shortened.
The other factor (which I hadn't considered) is the asymetry in the effective lengths of the lever arms that is introduced when the bar runs off the horizontal position - as you pointed out nicely in your diagram.
I believe that both of these have an effect and I guess it would depend on the relative magnitude of each to see how the overall roll stiffness would be changed by running the bar off horizontal.
One thing to consider is that the amount of body roll is generally fairly small, a few degrees, so the two sway bar ends aren't going to be splayed that far apart under normal cornering loads. In that case, L1 and L2 won't be hugely different from eachother so I'm not sure how significant the difference introduced by this factor will be.
I'm also not convinced that the statement that "the force on the outside is reduced" is the important factor here. The sway bar reduces body roll by introducing a torque along the roll axis that counters the torque that is making the body want to roll. If you look along the long centerline of the vehicle, this counter torque is produced by F1 and F2 acting on the attachment points of the sway bar, one up and the other down. If we call the distance from the roll center to the attachment points "d", then the (counter) torque would be something like (F1 + F2)*d, so it's (F1 + F2) that's important, not just the force on the outside. If one force increases and the other decreases, the counter torque could remain the same.
All interesting food for thought!