HP2100 core memory problems

Well, once again I come to the well for knowledge.....:) Seeking advice from those up on core memory I'm kinda lost on checking out the memory in my HP 2100. I have two 2100's, each with four 8K boards or 32k per machine. I had lots of parity errors so I lined up all eight 8K boards on the bench and reconfigured the memory controller for 8K total rather than 32K (test one board at a time). I put in one of the boards and ran a small memory test (not a diag tape, but a core memory test from the CE guide via the front panel).The core test has three "controlling" locations, one location for the first address to test, one for the last address, and then how many cycles per memory location. A cycle is defined as writing all zeros, read, and compare - then writing all ones, read, and compare all on a single location. Here's the symptoms. The full test of 8K takes about 2 hours (with a cycle of 3). On different boards it gets to different locations before the parity error halt. So far mostly at the middle or end, I don't recall it ever parity halting towards the beginning. I'll get a parity error halt on say location 012336 for example. I then manually go to the failing location and try storing different values in the location and reading them back out (all via the front panel). Most of the time this gives a parity error like you would expect. But - many times it doesn't. Then just for kicks I restart the diagnostic a few locations lower (like 012320 in this example) than the failing location. It fails right away usually (like location 012322), quite a few cells before the location which originally failed. This causes me great confusion, because it's actually now failing on addresses which previously tested OK just a few minutes ago on the previous test pass. Every 8K board I have exhibits this problem, but all at different locations. Supposedly both machines worked fine before they were put in storage 20 years ago, so these weren't picked up out of a junk heap or anything. Cosmetically they're beautiful inside and out. I've also tried switching the XY driver boards, the memory controller, etc. etc. but I can't seem to get anything stable enough to start intelligently swapping parts when every combination is bad. In case it matters, each 8K board is 17 bits per word (1 parity plus 16 data bits). Also, it isn't a complete failure - I've keyed in many other programs from the front panel and read in paper tapes (a suite of I/O tests and such) that all run fine so I know the system is somewhat coherent. I've checked the power supply for the memory cage and it's supposed to be 20 volts - my VOM came up with 20.48 ISTR, which is probably close enough I would think. Can anyone suggest a course of action or possible culprits in this situation? I suspect I can't see the forest for the trees anymore :) Thanks! Jay West