We've been having good luck with the dns currently on sources. I don't think I have seen the srvfail problem. The query waits on reverse lookups seem to have been a symptom of local ndb misconfiguration, though it would be better if dns coped more sensibly with that. As I recall, the nameserver(s) for the reverse domain need to have the relevant ptr records; pointing the reverse domain at the wrong nameservers (implicitly or explicitly) seems to cause confusion.
The main outstanding bug that I'm aware of is that aging resource records seems to corrupt some data structure and dns never recovers. There are two new control messages intended to help debug this: "target N" and "age" (see dns.c). But if one doesn't age resource records, dns can grow without bound.
