I recently had this problem with an IBM blade with AMD CPU. It is a 64-bit blade that I've run for years with 8 GB and 32-bit RedHat 3, which means the OS only saw 4 GB. I upgraded it to 64-bit RH 4 and started getting very similar errors. The research I found said the EDAC reporting in the OS appeared at some point in version 4, around 4.5.
That meant I didn't know if it was the 4-8 GB section or just the OS now reporting the problem, so I went to the BIOS to run tests. The basic tests showed no errors but the advanced tests that take hours to run eventually showed the problem. Rotating different RAM modules to different slots indicated 2 of my 4 were bad and that the test quit when it found the first one. Putting a total of 2 GB in the machine with 4 different modules made the machine run fine.
For years the machine wouldn't reboot without a complete power off, which wasn't a problem for a HA database server but I reported the problem too. IBM was nice enough to change my reported 2 RAM modules and the whole motherboard so I can't swear the RAM fixed it.
Bottom line: look really hard at your RAM. You might try a bootable RAM test, such as an OS disk. It appears the standard RAM test programs are 32-bit, so they won't test above 4 GB. A 64-bit OS disk might include one that will.