I guess one stumbling block is authors worried that people will second guess their space group assignment - but what you suggest is a good compromise on the road to full deposition of raw data.
I doubt that's really an issue; for as long as authors have been required to deposit (merged) structure factors, there has always been the possibility that everyone else can second-guess some of the details of their structure. �In fact, one of my colleagues wrote an entire paper on incorrect space group assignments:
And of course we have the multiple retractions that could not have happened without access to the data; it is generally agreed that this is a good thing. �Most of us instinctively appreciate that if our data can't endure detailed inspection by random colleagues, they probably shouldn't be published anyway.
The bigger problem, right now, is that PDB deposition is still an unpleasant experience and there is no standard mechanism for including unmerged data.