So you tried known good EPROMs in the broken board, correct? If the address bus shows deviations after start and that is always the same, wrong data must be read causing wrong addresses on the bus or the board branches into different code (NMI stuck or similar). Have a look at the data lines and do the same thing as with addresses (will be easier with a logic analyzer). Have you checked the address lines at the EPROMs? Same signals as at the CPU?