I think I might have figured out what is wrong with the 2732A ROMs .
My theory now is they have been over-erased.
I know already that they were used ROMs that were cleaned up and had new pins attached to them and the pins re-Tin plated. They break off with a small amount of force. They would have been erased too, probably in a large high power UV chamber to do a lot of IC's at once. A few of them also have a damaged "cell" here and there and won't program.
The higher than standard required program voltage suggests to me that there has been some global shift in the charge distribution in the die.
I was reading that it is possible to damage a UVeprom by over erasure, but the nature of the damage was not described. But since the difference between a 0 and 1 represents a difference in the amount of charge in a particular physical location with respect to another, and the process of erasure and programming represents having the charges in two different locations and it is now requiring a higher programming voltage, I think somehow they have been over erased. The behavior in the character generator, looking at the characters with lines of pixels coming and going, suggests that a number of locations in the ROM are acting as though they are on the border between a 0 and a 1 when the programming voltage was 21v, but not 25V.
Does anyone think this makes sense ? And does anyone know if the over erasure problem is permanent or reversible ?
I'm wondering now if this "damage" is reversible, or not. In that if I over programmed them a few times with bytes all 00 at a 25V voltage and a brief as possible erasure to get all FF's, if I could shift the charge distribution back the other way and get them to program and work properly with the 21V programming voltage that they were designed for. Could be an interesting experiment to help verify the theory.