Yes, I agree that the dynamic approach is by far the most interesting approach. But I still think, overall, that any watermarking is weak, even if we are only discussing the protection of an algorithm.

Suppose I have a great algorithm that I do not want some other company to steal. I embed some sort of dynamic watermark inside that would be extremely difficult to detect. Now my algorithm is protected, at least I can sue the company that steals it. But that is only true if the thief is lazy and just rips the algorithm from my code and drops it into hisr software. If the thief is careful, he will reconstruct my algorithm, rearrange some of the variables or procedures, make it a little more or a little less efficient, etc., and destroy my watermark.

The strongest technique to protect key pieces of code that I know of, is to encrypt those sections of the program. A popular piece of software that uses this method is Elcomsoft password crackers. The idea that they use is very good, but their implementation is very poor.