Trying to compare costs between BigQuery and Redshift/Snowflake is really, really hard because of their different pricing models. I tried to be very explicit about what assumptions I am making so people can make their own judgements. You are correct that a single-user sending queries to the warehouse is not realistic, but it should still produce an accurate price comparison. Here's my logic:
* Suppose I were to send 10 times as many concurrent queries to the same warehouse. That would drop the effective cost-per-query by 1/10, which makes BigQuery look 10 times worse.
* But suppose these queries now take 10 times as long. According to my formula, that makes the effective cost-per-query 10x larger, which makes BigQuery look 10 times better.
As long as the performance of the warehouse is linear-ish with respect to number-of-users, these two effects should cancel each other out.
RE: PS: Don't worry, I made that one up just for that test cluster, which is now gone :)