Ah. I think I misunderstood you. As I read you, the string itself (any random string data of a specific length known by the pike process as a string), however produced, would be used as to crash-and-burn on the generation of a [crypto-secure-object-by-whichever-name] whenever any kind of overlap of the two sets existed.
Now I think you are talking about some specific random-key-generation- function to crash and burn on generating the same output twice (when it manages to detect that having happened). The latter sounds like a potentially useful trait. The former did not.