Unclear behavior with option "lame-ttl 0;"

[Tom]

Hi
While testing the option "lame-ttl" with values 0 (disable) or any value 
higher zero on our resolver, I'm unsure, if I missed something (BIND 
9.11.2):

lame-ttl defines the time in seconds, how long a lame-server-entry 
should be cached and therefore not should be further asked (because it 
seems to be down). See 
http://www.zytrax.com/books/dns/ch7/hkpng.html#lame-ttl

If BIND recognizes a lame-server (written in the logfile), the 
corresponding server will not be contacted for authoritative queries. If 
I set the value "lame-ttl 0;", which means, caching lameservers will be 
disabled, then I would expect, that BIND will do "round-robin"-queries 
to all authoritative servers of a zone (includes the down-one). BUT: 
BIND still would have a notice of the lame-server (written in the log) 
and this server will still *NOT* be contacted for lookups.

I've tested with simple iptables-rules on my resolver, which are 
blocking outbound-connections to one or more authoritative servers of a 
zone for simulating the "lame-servers"-behavior.

Any explanation or hints for this (mis)-behavior?

Thank you.
Kind regards,
Tom