My experience is that all of these programs performance varies wildly with both type of image and even the specific image. I have had great success and total failures with each of them (except Photoshop, haven't tried that). Sometimes the reason is obvious, but in most cases I haven't been able to figure it out.
Manual merging can be very tedious, but it is the only method that always works. One reason is that all methods, including manual will have some discrepancies unless both the originals were absolutely geometrically perfect throughout there production cycle, from the original through how ever many generations of reproduction until you get the one you are working with. The human operator can recognize/decide what is important to him or her in the output image and do whatever is necessary to achieve that. For me that often includes such things as eliminating coffee stains, mis-spelled words and strange scintillation on lines which weren't perfectly straight and aligned with a digitization line.