502 errors with dead upstream servers & high traffic

[Fry-kun]

I installed latest nginx (0.7.23) with latest fair balancer and gave it ~300 queries per second traffic (60 upstream targets).
Some servers (1 or 2) were unavailable at the time, which is a very common condition for a cluster this size. I immediately started seeing 502 errors showing up in the log
The interesting chunk of the log lines, $upstream_addr:
my_cluster : my_cluster

The example is the most common cause of 502s, not even sure how to interpret it. When fair is not used, there are no errors - the failed servers are properly skipped by nginx.

I'm pretty sure that the cause of the problem is the dead servers, not the amount of traffic.. but mentioning both just in case.