BIND 8.2: 99% server load when lame delegation occurs (reprised)

[John N Dvorak]

In early May, Fabian Uebersax posted a question to the list regarding lame
delegation.  Essentially, any lame delegations to his name servers caused
his nameservers to refer back and forth to each other for a period of
time, driving CPU usage skyhigh (99%).

Cricket reponded, explaining the scenario:

>a asks b
>b refers a to b and c
>a asks c
>c refers a to b and c
>etc.

I don't have any other archived messages regarding the issue, and the May
issues are not up on isc.org yet. :-p I am trying to find an answer to the
configuration issue: how should BIND 8.2.1 be configured to avoid this
problem since lame deligation is inevitable?

This same problem is now occuring with our DNS system, which comprises of
3 authoritative slaves and 1 unannounced master.

Ever since the upgrade to 8.2 last month, lame deligations have caused
this looping. Examining the logs seems to suggest that just two servers
are involved (the two listed as SOA sources in InterNIC).  A asks B, B
asks A, A asks B, etc.  Am I missing something?  How could I have
misconfigured the DNS so that 'C' is now involved (if that is the problem
at all)?

Thanks for your assistance,

John Dvorak