Powerdns was vulnerable to the Kaminsky attack, but in a different way. It was actually easier to spoof the server due to its more actively dropping certain DNS packets. So while it did perform source port randomization, it was not totally immune to the attack either.
http://doc.powerdns.com/security-policy.html itself states:
All versions of PowerDNS before 18.104.22.168 do not respond to certain queries. This in itself is not a problem, but since the discovery by Dan Kaminsky of a new spoofing technique, this silence for queries PowerDNS considers invalid, within a valid domain, allows attackers more chances to feed *other* resolvers bad data.
Though it is phrased as "someone elses problem", in the DNS word of course nothing is "someone elses problem". DNS servers are chained in hierachies and one problem somewhere leads to problems elsewhere. DNS is all about protocol compliance to ensure interoperability. With the "someone elses problem" approach, we would have had no "reflection attack" and "amplification attack" problems either, it being "someone elses problem". Despite the nice phrasing, powerdns caused cache poisoning problems as a result of the Kaminsky attack that needed to be addressed.
In general, I have a problem with bug reports and changelogs writing things as "improved error handling", "made more robust" or "add security to" which are too often used to hide the real security impact of certain bugs. DJB's policy of "it is not my bug to fix, because it is an operating system bug" is also completely bogus from a system administrator point of view who still ends up with a security problem.