Hello jazzyj,
You can use single GCP with high accuracy in addition to the low accuracy camera locations.
Also, if you know the shift to be applied, you can add the same fixed X, Y and Z values to the source coordinates of every camera (it may be automated with Python) and apply the referencing.