Initially I used the old arms, effector and carriages after upgrading the frame and changing to linear rails. The auto calibration deviation (using 9 peripheral points, 3 points at half the build radius and the centre) halved from about 0.1um to 0.05um. But I hadn't yet adjusted the Z probe trigger height corrections to account for the effector tilt being lower than before. Had I done that then the deviation might have been lower still.
I'll report the new auto calibration deviation when I have a working Z probe again.
I upgraded from 2020 to 2040 extrusion because my delta is 1m high and I think the extra rigidity will probably be worthwhile, especially for tall prints. When I first built the printer, the metal corners were only available for 2020.
I have a draft blog page about this printer build along with a detailed and costed BoM.
The lowest friction possible between the cups surrounding the magnets and the balls also helps. This can be achieved with grease, or use of delrin/acetal cups.