This is an interesting puzzle! If the problem is happening on multiple Reloaded boards owned by multiple people, including ones with all original components, then it suggests there's something different about the Reloaded board as compared to the original SE/30 board. I believe the Reloaded board is not a direct copy of the SE/30 board - it has the same logical connections but the traces are routed differently? You may have a situation where there's more parasitic capacitance or coupling between adjacent traces than the original board, and problems appear or disappear with tiny changes in the supply voltage, temperature, or by replacing a chip with an identical chip whose properties are slightly different due to normal process variations.
That would be a very difficult problem to solve, and you'd probably need to watch analog-level signals with a scope to diagnose it. If you can ID exactly what's going wrong, perhaps you could add or remove extra capacitance from a problem trace to compensate, or cut a trace and replace it with a patch wire.
It sounds like you already know that wrong values are being written to RAM. Next would be trying to determine whether it's a spurious write to RAM when no RAM was meant to be written, or if it's a valid write but it's written to the wrong address, or if it's a valid write and written to the correct address, but the data value was stored incorrectly.