Why so many buildfarm errors with contacting "git.postgresql.org"?

I've noticed that the frequency of nonrepeating fetch failures in the
buildfarm seems to be a lot higher with git than it ever was with cvs.

A typical example is today at:
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=sloth&dt=2010-12-07%2018%3A30%3A01

fatal: Unable to look up git.postgresql.org (port 9418) (Temporary failure in name resolution)

and similarly two days ago:
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=colugos&dt=2010-12-05%2021%3A05%3A56

11 days ago:
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=coypu&dt=2010-11-26%2021%3A05%3A02

28 days ago:
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=mongoose&dt=2010-11-09%2011%3A45%3A01

45 days ago:
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=polecat&dt=2010-10-23%2018%3A49%3A59

That's just name resolution failures; there are a similar number of
Git-stage failures due to connection timeouts. The problem appears
to be getting worse with time :-(

Is there any difference between the network connectivity of
git.postgresql.org and the old anoncvs server?

regards, tom lane