Strange behavior in the RICM PDP-8/L

I am hoping that one of you might be able to shed some light on a strange behavior in the RICM PDP-8/L. It looks like we have everything working OK except for a strange behavior when the processor is run at full speed. Single-stepping Instruction Test 1 works OK, so at least the processor logic is mostly functional. We tried just running Instruction Test 1 at full speed, but it halted at 0501. We looked at the code in that area and found that the contents of location 0500 was all zeros. We loaded a little program consisting of: IAC, IAC, JMP .-2 and found that it would replace the first IAC with zeros. More experimenting showed that if any of the address bits 6-11 were on, the program would work OK. If you run the processor at full speed through an instruction at xx00, that location is replaced with zeros. We swapped all of the G221s, G228s, and G224s. None of the module changes affected the strange behavior. We swapped the M617 in slot A9, but that didn't make a difference. The read/re-write current waveforms look OK for all addresses. At this point we really don't know what is causing this behavior, so any ideas would be helpful. -- Michael Thompson