Having read it I love the approach - it's very similar to how we handled data at Common Crawl / Web Archive (WARC) files / inspired by MapReduce + object storage interactions. I'll keep it in mind for the next opportunity :)