I think to have found something important!
As i said i would try to isolate the problems as simple as possible.
So i took the object that triggered me.
Here you see it duplicated 5 times, but all slightly turned to get different lightmaps.
But also the object settings are a little different.
The only door with a wrong lightmap is number 4.
and i think to know why!
First tell you the diffrences in the objects.
Door left (number 1) contains 4 objects with 1 material each.
a left door
a right door
a wooden bow above the door
and the stone entrance around the whole.
The second door has only 3 objects
a left door 1 material
a right door 1 material
and the entrance with bow 2 materials
The third door is exactly the same as the second door, only slightly turned.
So my conclusion is here, that 2 materials in one object gives no problem!
The third and the fifth door are the same as the second door only different positioned and turned.
You see that both got two different lightmaps , they differ from the second door.
Door 4 is the interesting one, it has shared materials with door 3
Because it is the same angle as door 5 , it should have been lighted as door 5
But it looks like it took over the lightmap from door3 but the preview lightmap tells me a different story.
In a closer look you notice that it is also a bit darker then door 3
I predict that is the addition of lightmap 3 with lightmap 5!
Korman goes through each lightmapped object, selects all the vertices, applies uv mapping as you describe here, and switches to a temporary lightmap texture. Then Korman bakes batches of objects that have the same layer selections to those temporary textures.
Referring to this quote, you combine the materials into batches, i think you forgot that the combined batch will have overlap again for each object in 1 batch,
because each object was projected and completely filling the same sized area of the uvpack frame, inside the editor,
and because the uvlayout for door 3 and 5 are the same , they have perfect fitting overlap and add to each other .
So when it would have been a different object , it will result in a mess.
Probably the solution is that you have to perform a new "crtr P" to repack all those overlapped uv maps again in your combined, temporarely material layer.