I wrote the post in about half an hour and stupidly enough, I did not think about UUID v1. I knew about v1 (from the Melissa virus case), v4 and v3 & v5 (which I once used to generate deterministic UUIDs). I just didn't link my case to UUID v1 because I've hard-wired UUID with UUIDv4 in my head.
Here are two observations from the discussion on HN.
First: over time, the first characters of UUIDv1 are still uniformly distributed.
The only reason I had the same prefix was because I filled in all values with one UPDATE query when I initialized the column. If on the other hand you use UUID() whenever you create a new row, the first characters will be distributed evenly.
Second: the different UUID versions should be more clearly distinguished in general. The Zen of Python has a line:
"Explicit is better than implicit."In following this, I think it's better not to expose any UUID() function, but only explicitly offer the typed versions UUIDv1(), UUIDv4() etc. This is already what most libraries do as far as I've seen.