I was able to consistently reproduce the timeout on my Gentoo buildbot
running in a vserver virthost.  After some experimentation and perusing
the related issue 3892, I discovered that by increasing the timeout I
could get the test to consistently pass in my buildbot.  However, it
does occasionally produce the DB_LOCK_DEADLOCK traceback in one of the
threads. (Two times out of about 17 runs in a loop).

I've checked an increase in the timeout for the two replication tests
into trunk in r76116.  In some of my experiments it was clear that the
base replication case is much quicker than the ReplicationManager case,
so I made the ReplicationManager timeout longer.  I'm hoping this will
fix the line 315 failure in the windows buildbot, since it does
sometimes pass.