That's compression ratio estimation, not compression level (or parameter) estimation. I thought you were talking about the latter, as the former is virtually impossible (some archivers do sampling as you suggested, but it's only an opportunistic heuristic).