I have actually done something like this in 2006... but I was dedicated to loss based Algo, where you could recover the full compressed data but, only under full storage of the offseted data. Basically making the offset very large, in such way you can compress lots and diverse amount of data, while only holding the offset (very big one).

Study cases where, for example... having 1TB of offset, would allow me to compress Petabytes... into Megabytes.