Using a carefully decollimated primary, the focuser axial error is magnified 2X, no compromises. Reducing the primary mirror after correcting the focuser axial error, the error can be read to 1/8 of the autocollimator pupil diameter, or 0.015-inch. I can usually read a Cheshire alignment to about 0.01-inch, and maybe a little closer with a good Barlowed laser. For high magnification performance, my f/4 primary mirror axial alignment needs to be in the vicinity of 0.014-inch, which is pretty close to the resolution of the autocollimator and well within the resolution of my "Barlowed" (1mm aperture stop) laser. Since I use both tools, I'm always satisfied with my alignment accuracy.
...Although this is probably just another academic point, considering the calculations Jim presented earlier on this page, it does seem to me that the dissappearing compromises the 2x-4x-6x-8x accuracy claim of the AC at least a little. If that didn't happen its accuracy superiority to a cheshire would be more convincing. I'd buy one, anyway.
The real power of the autocollimator lies in its ability to show persistent axial errors, magnified as much as 8X, when a less reliable tool is used (i.e., an "economy" laser), or when the user's alignment skill isn't quite up to speed compared to his performance expectations. With more reliable tools (including the user's skill, experience, and knowledge), the autocollimator's accuracy may not be necessary for final axial alignment, but it can still be used for redundant verification.