[CentOS] Obscure behavior with filesystem update

I've been getting weird results with the recent filesystem update. It 
appears to be hanging at the transaction test stage endlessly. I can close 
the ssh connection and the update still goes on. When I use screen I 
cannot reattach after logout because the screen still shows as attached 
and I have to use -D -RR to get back to it.
Sometimes it succeeds eventually after half an hour, sometimes not. Those 
that didn't I have been able to get finalized with screen.

An strace shows an endless loop thru the man directory from a-z and then 
starting again, like this:
stat64("/usr/share/man/aa", 0xbfc82584) = -1 ENOENT (No such file or 
directory)
Anyone else seeing this?

Also, there is weird behavior concerning the list of available updates. 
Yesterday I had a few servers that didn't show apr, filesystem and mysql 
as updates although all were using the same mirror (mirror.atrpms.net). 
Today they have shown up. Is it possible this is a problem created by 
"mirroring a mirror"? e.g. mirror.atrpms.net having its own mirrors that 
were at different stages of update?

Kai