Why would that make a difference unless some sectors have higher/lower error rates than others?

For a fixed bit error rate, making your typical error 100x bigger means it will happen 100x less often.

If the typical error is an entire sector, that's over 30 thousand bits. 1:1e15 BER could mean 1 corrupted bit every 100 terabytes or it could mean 1 corrupted sector every 4 exabytes. Or anything in between. If there's any more detailed spec for what that number means, I'd love to see it.