Our point2dem --orthoimage workflow is very tied to PC.tif and L.tif being precisely one-to-one. The only kind of manipulation that will work is, I guess, to move PC.tif with some alignment transform, which won't change its size or structure, then use point3dem with that. Once you move from PC.tif to DEM, you can't go back.
If you have DEMs, the only way to create orthoimages is with mapproject, and that is very much the suggested approach.
The pc_align command is not sufficiently accurate for you to get DEMs that you later merge and for that to result in seamless ortho mosaics. Any small rotation or translation error will result in a seam.
The only way we were able to do "full NAC stereo site's mosaic" is to do bundle adjustment with lots of matches of all relevant NAC files. Then these are moved as a group to align to LOLA, say. Then you make joint stereo DEMs, from some pairs of your bundle-adjusted cameras, then you mapproject onto that each image. Then you merge the ortho images with dem_mosaic. One has to be careful with areas in shadow. We have some options with dem_mosaic for thresholdiing, doing max, etc.
Some minor inaccuracies may still result if there is jitter. Any stereo pairs should also have a good convergence angle (say no less than 15 degrees).
Global registration can be quite hard to get it right. We did it on a very large scale, say 50 x 40 km at 1 m/pixel, for LRO NAC with more than 3000 images, but it is a pain.