>This is why good general purpose memory tests are largely a myth.

Indeed.

It is also worth noting that the very act of *testing* the stuff
can make it *work*.  There was a batch of Seimens DRAMs that went
into MicroVAXen, where the refresh logic was broken.  VMS had an
idle-loop page-scanner that kept everything alive.  Testing the DRAM
showed it good because testing kept everything alive too.

Chris