I believe the main pitfall of this approach is related to what the author of the og article mentioned around r/w and query latency from S3. I like the old Twitter eng blog post on their distributed kv store called Manhattan (https://blog.x.com/engineering/en_us/a/2014/manhattan-our-re...). For any serious real-time data, we probably still need some semblance of fast R/W storage, but analytics workloads or long-running tasks can def benefit for cheap blob storage.