I'd recommend comparing the BIOS versions with something like: "dmidecode | grep -A4 BIOS" to see if there's a correlation between the ones that got the UP kernel instead of the SMP kernel.
It happened to me too. I figured I was the one made an error and manually fixed the "broken" ones. Mine were a mixture of HT-enabled P4's (which would normally run the smp kernel) and some dual opteron machines. These are all rack servers configured identically so I'm not sure what triggered the behaviour in some of them without affecting them all. Wierd.
-Jay