ah! got bit by this and had to workaround serving stuff from s3. nice to see there was an actual problem and wasn't just some issue with our stack.