Late to reply but I decided to have a go at it. The end result at least looks convincing, but there are a few assumptions. Have a look in test Arm.blend (961.2 KB)
Pay extra attention to bone Arm.Center
. This is a “fictitious” bone so that the IK solver gives the upper and lower arm an equal angle towards the joint. I just assumed that this bone would have the same length as the center-center distance between the furthest joints of each of the three “paths”. This could probably be proved via trigonometry, but it at least looks ok. I ran out of ideas for the top joints, so I have them just track a bone “projected” on the opposite lower arm. Again, no idea if this is supported by the mechanics, but it looks ok.