The obvious answer is bad memory.
At the board level, yes. Deeper, it could be bad memory bits or bad
memory decode.
Yes, one of the standard early PDP-11 memory tests is the "no duplicate address
test".
I should say that the memory board is not _completely_ whack -- it is passing the rather
thorough MAINDEC ZQMC, a 0-124k exerciser with multiple pattern/sequence tests which also
kicks around the KT11.
That doesn't rule out the possibility that there is a lurker in there not covered by
the DEC diags. But if there is, its something subtle...