Tobias, thanks for the information. In the BW world, this would be something like deleting any change pointers/delta initializations during the population of any unneeded data, then turning them back on at a later time. I would see this as a viable option vs. the second point, as in some processes the deletion of data should be replicated.
Do you have any reference material to the option you mention in your first point?
Thanks,
Justin