The point is that the lockbox is the TPM that, on paper, is supposed to be unbreakable. In practice, sometimes it can still be broken with physical attacks (like side channel analysis or fault injection, or even simply snooping the communication between the TPM and the rest of the system with a logic level analyzer), despite that it should be designed to be hard to break even with such attacks.

If the TPM is properly designed and manufactured, and the software relying on it is again properly designed and implemented, then it would be perfectly secure. The problem is more the difference between the theory and the real world; the flimsy lockbox analogy doesn't hold.