Comment by md_

How does that compare to just doing compression on a sample of data? :)