> From my perspective, the ideal approach to keeping older backups at
> a lower density than new ones would be writing code to be able to
> merge reverse-diffs, such that two historical diffs could be
> replaced with only one containing the changes from both -- and
> excluding the intermediate changes from an item which changed
> multiple times between the merged set.
> Anything else requires storing extra or redundant information
> beyond what is currently done now.
This would also suit my needs. In fact it is better than my idea.
My proposal would have resulted in many full backups (with no diffs)
for the older data, but may have been easier to implement.