Sounds like a missing speed path in the test coverage, since the problem is not present at 800MHz. In fact it sounds rather like the P3-1.13GHz problem.
Yes it does but in fairness, there are more speed paths than can possibly be tested. There is no know method of quanifying speedpath coverage at test. It is an empirical process.