I only just now came across this post. I had observed the same problem a couple of months ago. Polar alignment got progressively worse with each run of the alignment module.
At the time I thought I was doing something wrong, so I went back to polar aligning with my polar scope, which works very well at up to 300s exposures at 1600mm with perfectly round stars, so I am obviously getting very close to the NCP.
Haven't gone back to using the polar alignment module since, but would very much like to as soon as this has been fixed. I believe I was using the nightly Ekos build at the time.
Mount: iOptron SmartEQPro+
Imaging scope: WO-Z61 (not used for alignment)
Guidescope: Orion miniguide scope (30mm/130mm) (not used for alignment)
Polemaster: camera mounted in polar alignment scope tunnel, in the rotational axis of the mount (used for alignment to eliminate deviations caused by flexure)
I saw exactly what was described by Gonzo: With every iteration of the alignment procedure, my deviation approximately doubled.
I then tried it again using my imaging scope. Results were the same. Deviation doubled with each iteration and it made no difference whether the mount was rotating clockwise or counterclockwise (i.e. East or West)