U: isn't very useful any more since the advent of higher difficulty shares.
If you mine at 2 diff only, then you'd double U: to get the equivalent 1 diff value.
However, you don't usually mine at only a single diff value ... it changes at least once when you start mining
(and on some pools it changes often)
WU: is way more useful in that respect ... however it includes rejects etc.

U: is still shares per minute.  The WU: is a good estimation what you are being credited by the pool.  In my case they are pretty close since all of my work is diff1.  Still I think it's good information to have your rate at your fingertips.

I just have no idea what the OP was getting at so I was just attempting to explain what the U: meant.

That sort of error (not specifically that, but it's one possibility) would happen when you run out of ram... the question is why are you running out of ram.

Nope. It can only happen when it runs out of some kind of resource. It doesn't mean the machine itself ran out of ram, but the OS refused to give it any more.

Perhaps it hadn't disconnected but you were sending lots of shares and cgminer was simply waiting for the response to all of them and btcguild wasn't delivering. cgminer needs to keep a record of shares submitted until it gets a response. I've been meaning to find some useful way to throttle how many records it stores to prevent this from happening.

Could be unrelated. Was there any other message? It would tell you if the data was corrupted which is more likely if they all had it then... and wtf is <1MHash? Cgminer doesn't run on any hardware that slow.

I want to run 1 bf single on a separate pool. I tried the "-S COM9" command, but it still starts all the units. I thought I had done it that way in the past. Is there still a way to do it?

On the next build can "device" : "0,1,etc..",   
get fixed for conf files?
Ive tried the above syntax and also listing "device" several times w/ one value for each gpu...it just uses the last one. This has been a bug since very early versions of cgminer and surprised its not fixed.
Unless Im doin the syntax wrong....

cool, yeah I know its a bug now. I ran cgminer, manually disabled a gpu, wrote the config from within cgminer to new conf file, told cgminer to use that conf and it ignored all "device" entries except the last one. So, all devices got devices cept last one =(