Hi Götz
One smallish project I tested did not allign all images into one component but 2, even though there were some very distinct features (handwritten numbers) present on images in both components. I tried adding control points to help with the alignment. Maybe I did not distribute them enough over the surface, which is due to a texture with few (human) recognisable features. After doing that, the 2 ccomponents were merged into one, but now there were faulty parts in the mesh. My guess is, that the control points were too close together in too small an area, so that the natural placement errors multiplied and sreweed up the alignment. I could swear I saw an option somewhere that said something to the extent of “control point priority” (which I thought meant that I can define, whether control points are absolute or just helpers for alignment), but I cannot find it again for the life of me. Did I dream that???
Yes you guess right, its best to set the CPs a bit spread over the image
take a look how i do it https://www.youtube.com/watch?v=d8naLEtLqDY
Have seen it good, not dreamed about it 
There is a option to set higher “priority” or better WEIGHT. Just click on CP and get a small info on the point and there can set it weight of the ControlPoint
And a follow-up question relating to that:
When I have defined sufficient control points and hit alignment again - is that a completely new process or just a refinement? That is interesting in terms of processing times.
- if have 2 or more components, and add CPs then need align it. But as have seen our alignment speed is quite fast even with 1000´s of images.
- When get bad alignment even with the CPs try delete all components in the project and align it again.
I don’t quite understand the components and why they remain after alignment - do they work similar to the chunks in photoscan? Does that mean that the second round of alignment is rather equivalent to aligning chunks?
Sometimes results are separated to component. but we use the idea much deeper and mostly for large datasets.
Will prepare a indepth explanation how to work with Components and how to use to its maximum potential. its a very powerfull tool and easy to manage, but it can look complicated on first look.