Black Hole
May contain traces of nut
If tolerances are shown to be the problem, I would posit timing discrepancies as the most likely candidate.
For example, the drive might require a certain period to elapse between actioning one command and being able to action another, and that period might vary slightly in the production batch, device temperature, what it did last, time of the month... If the old kernel issued commands well outside that "dead" period, but the new kernel is right on the margin, some drives will have a percentage probability of appearing to malfunction, but not necessarily be outside their manufacturing specification.
I have absolutely no data to support this as fact, I am just illustrating the sorts of thing that can happen in hardware and software development (and the interaction between them).
For example, the drive might require a certain period to elapse between actioning one command and being able to action another, and that period might vary slightly in the production batch, device temperature, what it did last, time of the month... If the old kernel issued commands well outside that "dead" period, but the new kernel is right on the margin, some drives will have a percentage probability of appearing to malfunction, but not necessarily be outside their manufacturing specification.
I have absolutely no data to support this as fact, I am just illustrating the sorts of thing that can happen in hardware and software development (and the interaction between them).