From a quick look I didn't notice any mention of columnar storage. I would be very skeptical about any claim that a DB without columnar storage "can work extremely well as a data warehouse".
The fact is that for large datasets scans on denormalized fact tables parallelize well, which means you can (a) offer stable performance and (b) scale more efficiently. This is important for use cases like web analytics, where users play around with different dimensions and measures but still expect consistent response. Note also, dimensions for things like Year, Month, Week, and the like compress absurdly well. It is often way faster to scan these values than to join them.