There could be other possibilities:
1. The chips are already close to the TDP when operated at the maximum allowed temperature and voltage. In that case, even if they'd work at faster speeds, they'd consume too much power when run faster.
2. There's a lot of margin built in because Intel expects significant slowdown of the transistors due to aging effects (NBTI, hot-e in particular), so the chips will OC really well for a few years and if kept at low voltages and temperatures.
3. The chips won't last long enough if clocked higher. If the metal wiring in some circuits wasn't thick enough to carry large amounts of current, the frequency would have to be limited to prevent the chips from dying after just a few years. Alternately, the voltage might be limited by wearout effects like TDDB and operating at a higher voltage would result in unacceptable failure rates after a few years (even though they work fine for a while).
4. There are critical paths in logic that isn't normally used. For example, x86 has a feature called "segments" which aren't used by modern OSes. If a critical path occurred in logic like this, most people wouldn't notice failures.
5. There are critical paths in logic that doesn't break things. For example, if the branch prediction logic doesn't work properly, the CPU would still work fine, but slower.
4 and 5 seem unlikely to me.