The ML module implements the method faithfully, so you're asking whether the method is dependable. For my own part, I've been very satisfied with DotTune. When I first upgraded to the 50D, I painstakingly made lots of test shots and manual adjustments to figure out good AFMA values for my lenses. Then I found out about DotTune, and I was pleased (but frankly also a little irritated!) with how easy it was, and produced AFMA values very close (<~1 tick difference) to what I had already determined.
Now, there are definitely times when DotTune misfires and gives a bad result, but I chalk that up to user error, such as not setting the perfect focus initially or accidentally bumping the camera when switching to manual focus. Running through the procedure a few times can help confirm whether your value is trustworthy.
I suspect that those that see consistently incorrect AFMA values from DotTune are making some mistake in the setup/procedure and not realizing it.