I have a Mamiya AFD and a back with 9-micron pixels. I have not noticed issues with AF accuracy with the 55-110mm zoom, but those shots tend to be handheld so I'm not expecting perfection.
Manual focus on the screen is definitely off, so I'll need to shim the screen a little bit. But manual focus using the focus confirm system is accurate, and to be honest, because of this I haven't really been too bothered about the focusing screen's obvious inaccuracy. The sprung clip system for holding the focusing screen is a definite weak point, with too much potential for wiggle. Older Mamiya 645s (like my 1000s) with their "top loaded" screens were far more rigid, and adjustable to boot; the Hassy V sounds similar. Maybe the Contax and Hassy H are also better in this regard; I have no experience with them.
So when I manual focus, I either use the focus confirm system (for speed) or I use 100% zoom on the back, especially if I'm using a tripod and can repeat the shot.
When I set up for astrophotography, I start with the lens as far over as it will go (at or past infinity) and move the focus back in 6 or 7 very small increments (it helps to stick on a finer tickmark scale beside the focus scale), inspecting the shot at 100% each time. A bright star at 1/6 second exposure, lens wide open, mirror locked up, self-timer on, does nicely. With a lens like the 200/2.8 APO wide open, this process allows me to mentally construct a simple "focus curve", as the star comes into maximum sharpness and then out of it again. After about 2 minutes of this testing, I set it to the position that gave the best focus and I'm good to go for hours of imaging - weather permitting! I would generally re-test focus later if there has been a substantial change in the temperature since I started.
I think that this "mental focus curve" method should work just as well for discerning landscape work etc. It's one great advantage that these backs confer over film.