And now I see what these text IDs are used for:
the lsx dialog file stores texts with unique IDs and nodes (the whole structure and settings). The IDs are used to find the right text for the nodes. So using the same ID in different nodes for the same text shouldn't be a problem but even be more efficient! Now there's just one text entry for 200 nodes instead of 200 text entries for 200 nodes. If my nodes had different texts this wouldn't work, of course.
