Transform between map and nav graph for big maps

[Fellfalla]

I am concerned about the precision of the current way the transformation between maps and nav graph happens for use cases with target maps that extend over several hundret meters in one direction. That means any small change in the curvature of the localization map on the robots over time (e.g. due to livelong mapping) or initial errors might cause big errors at points far away from the origin. At least that's what I'm concerned about based on what I know so far, as the RMF highway points are connected to the map via a single transform.
 
Is it an actual hazard and if so, are there any ideas to mitigate this drawback?