The addresses where it fails are not always consistent, which leads me to believe it's some sort of noise or timing problem. If there were some serious trouble on any of the lines, like an open or short, I'd expect there to be a lot more problems - like 1 in 2 or 1 in 8 reads/writes failing instead of about 1 in 32. The only pattern I could find was that it tended to goof up more on addresses which had more "set" bits in the low-order byte of the address; xx00..xx70 or so would almost always work, and xxF0..xxFF would get three or four problems. I suppose I could log the errors by having it spit them out the serial port.