Thanks for sharing your tracking results using DEC arc modeling, this is very useful information for others looking to try it. Looks like 3 arcs works very well, do you know if 2 arcs works? Rolando is correct that it all comes down to the tracking error in arc seconds per minute x the plate scale:
Image Drift (pixels) = exposure time (seconds) x tracking error rate (arc seconds per second) / plate scale (arc seconds per pixel)
The greater the drift in the exposure (pixels) the more elongated the stars will appear.
Which means you can reduce stellar elongation by: a) reducing the exposure time, b) reducing the tracking error rate, or c) increasing the plate scale.
The new DEC arc modelling method does (b) which means for a given drift (star roundness in an image) you can now increase the image exposure time. The equation also shows that putting a focal reducer into the optical path (increases the plate scale) can also reduce the image drift for a given exposure time and tracking error.
My plate scale is 0.5 arc seconds per pixel so you can see my system is about 4x more sensitive to tracking error than your system.