The culprit is the grafting in of IPv6 readiness in man 2 getaddrinfo. This is the way of the future, so there is no fighting it on a long term basis, but tactically having a means to be IPv4 only is appealing for people just wanting to work in the older address space. The TL; DR uptake of that man page is that in a properly functioning system, name resolution answers under IPv6 are preferred, and only if not available, does one fall back to the older IPv4. But this places a premium on IPv6 actually working when present. We've shipped a full native IPv6 setup for customers at PMMan for a couple of years ago, but I assure you that we had some head-scratching as we rolled it out, and found customers using tunnels from HE or SixXs were also leaking advertisements to other VM's. We added rules to filter out the false traffic after a bit of tracing with TCPDUMP
I have blogged about it before when IPv6 link-local address resolution (the ^FE family) was confusing distcc under Debian a couple of years ago. There are links in the CentOS wiki for approaches on disabling IPv6 traffic, which vary between C5 and C6
That last mentioned article has an outlink to a bugzilla ticket that offers food for thought. It mentions in passing that one can direct a nameserver to NOT deliver IPv6 results with a fairly simple tweak
Another option is to add this to /etc/sysconfig/named:OPTIONS="-4"
... so, ... it should be possible to set up a local cacheing nameserver on the localhost, configured to NOT return IVv6 content, and so workaround the issue. This smells sort of 'hackish', but it would have the benefit of being a single method that should work in the general case and not be tied to any particular kernel version, or other variable
