Intermittent buildfarm failures in sequence test

Have a look at
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=raven&dt=2011-01-05%2001%3A30%3A12
http://buildfarm.postgresql.org/cgi-bin/show_log.pl?nm=currawong&dt=2011-01-06%2002%3A30%3A01
I recall seeing a couple of similar failures in the past few weeks but
can't dredge them up at the moment.

The test case is pretty simple:

CREATE SEQUENCE foo_seq;
ALTER TABLE foo_seq RENAME TO foo_seq_new;
SELECT * FROM foo_seq_new;
SELECT nextval('foo_seq_new');
SELECT nextval('foo_seq_new');
SELECT * FROM foo_seq_new;
DROP SEQUENCE foo_seq_new;

In the failure reports, all the SELECTs give the expected values except
that "log_cnt" in the last one is 31 instead of expected 32.

Anybody have any idea what's causing that? I can't avoid the suspicion
that this is a consequence of some replication-related hack or other,
but I haven't been keeping close enough tabs to guess just what.

regards, tom lane