Were different power supplies tried? Or the original one moved to the second machine for testing? I'm thinking about /pfw power fail interrupts that when handled see power as ok again, but that monopolize interrupt handling time...these could be faulty /pfw signals from one bad supply, or good ones from two different good supplies that happen by their design not to be able to handle spike or dropout electrical noise unique to your lab...

Is there an unusual lab electromagnetic environment that could cause the normally 32768 Hz crystal oscillator to run in sync with some slower external humungous field instead? Like an NMR or MRI setup?