And with my tinfoil hat on, I'm saying it's probably driven by requests from three-letter agencies. We already know they scan the covers not just for automated processing (which as you mention wouldn't require saving the image after the OCR is complete), but to also provide it to law enforcement.
If public pressure is forcing those agencies to be more forthright with what data they're collecting and storing, it conceivable for them to push that responsibility upstream. If the USPS is going to be imaging and storing everyone's mail, there's no need for other agencies to do so. They can just access it on the fly.
Arguably, the government would save money with this approach as well as I'm sure all three major agencies would want a full copy the data set.
If you have a substantive point to make, make it thoughtfully; otherwise please don't comment until you do.