I found a (possible) workaround that helped in my case...
It appears to me that the incremental admin import has some problems with multiple node csv files that are different in structure and carrying elements of different labels (mixed within each row of the node csv files and mixed within start and end nodes in the relationship csv file).
All possible options I tried (with or without grouped ID spaces, label information within ID column, group labeled relations csv ID column entries etc...) nothing worked. I always ended up with more or less helpful error messages.
My solution to get this thing going:
I created one single node csv file out of all my initaly different csv files. Having different csv column structures within each initial csv file, I was forced to create as many unique columns as needed and left the row cells empty for each column that was not applicable.
So I ended up with a huge sparse matrix like structure within one csv file.
The only requirement to go down this path is that each column has a unique type (which is the case for my import) to set the import value right during the incremental import run.
Then I created one simple relationship csv file as defined for the admin import.
There is one more strange thing that needs to be done to get this whole thing flying. The incremental admin import requires "node property uniqueness constraints" to be set. Otherwise this thing will not even start. But what constraint should you set if all this stuff is now cramped in one single file with multiple labels?
The answer is: it doesn't matter. Just pick one. Preferable one for a label that is in your import so you don't end up with an extra "chip" in the "Nodes" section in neo4j Browser. But I noticed that in the end it doesn't matter. You could even pick a name like "Foo" if you like. It appears that the import just looks for the existence but isn't doing anything with it.
The only thing that is required is that the label for which the constraint is set is added to the first column header in the nodes.csv file as id:ID{label:Foo}
As long as the :LABELS column cell in the nodes csv is set with whatever label is applicable for the specific row the correct label is set in neo4j during the import.
One more thing: set "--ignore-empty-strings=true" to avoid processing the sparse cells of the "matrix".
Then the whole thing worked like expected and incrementally imported all the csv data into an existing neo4j.
It is definetly a strange behaviour. Maybe someone from neo4j can turn some lights on this issue. But for the time being I'm happy to be able to import my stuff into an existing DB using the incremental Admin update.
Krid