VME MultiProcessor Startup problem - Any suggestions out there??

Real simple system but an interesting problem.  Master slot is a old
Heurikon Malibu R3000 running a ram test to the first 100K words of a
VME Ram card (Micro Memory 6260)located in slot 2.  VMETRO in Slot 3. 
Power PC MVME 2300 running in slot 4 writing to the same ram board but
at a different 100k word block.

  On startup the R3000 reports errors in reading back the data it
wrote to ram...perhaps 1 error in every 1000 accesses.  Power PC never
reports an error.  There are times....like one in 3, when I fire up
the test and the system will run these tests without error over the
weekend!

   I can run the ram tests all day long in stand alone mode so it
doesn't apper to be the simple test algorithms I implemented.

  I have swapped the R3000 CPU and ram board.  I have swapped VME
chassi.  I only have one PPC board up so I am stuck with that for now.
 I suspected bus loading so I moved all the boards left towards the
master slot.   Individual tests on the power pc show it never reports
a failure on any ram test I run on it.  It is always the R3000 running
in the master slot.

  Right now I am ready to break out the voodoo beads and rattles.

ANy suggestions are most welcome!!

Candice Uhlir
Northrop Grumman Space Technology.