ngx upstream round robin bug?

[defage]

I use nginx as a load balancing for backend web server.
 if this upstream has n servers, and when the number of backend web
server can't connect(502 or down) be equal n-1, nginx will response  502
or 503 status to client.  
I did some test, only in the use of directer like:
    server 10.10.1.2:7003 max_fails=NUM1 fail_timeout=NUM2;

within NUM2 second time, send request times > NUM1, the (NUM1+1)st 
resquest will response 502 or 503. The above will be happened.

nginx version:1.0.11, centos5.4 x86_64
this is my conf
upstream backend {
    server 10.10.1.2:7003 max_fails=10 fail_timeout=100;  #normal
service
    server 10.10.1.3:7004 max_fails=10 fail_timeout=100;  #had down
    server 10.10.1.4:7005 max_fails=10 fail_timeout=100;  #had down
}

location / {
    proxy_pass        http://backend;
}

the access_log when error happened like:
1324630768.256 cas.sdo.com
/cas/login?gateway=true&service=http://newkoo.com 502 10.10.1.4:7005,
backend 125.64.219.39, 10.129.1.230 0.002, 0.000

Posted at Nginx Forum: http://forum.nginx.org/read.php?2,220415,220415#msg-220415