My personal experience with "wounded" chips is that they are much more likely to fail than a totally good chip.

For example, consider a spec of dust on a chip which destroys the memory bit lying under it. It may also destroy half of the adjacent bit line. Using this chip in an application that avoids the bad bit could cause a later failure when the damaged bit line finally opens up.

Chip testing is an art. You can't check everything. So the best methodology is to throw away anything that might be bad in any way. Silicon is cheap. Returns aren't.