You don't save them for the same reason that you don't necessarily save the state of the simulation after every state change. You're generating a lot of these things, conceivably billions and billions or more. Even if you do save that kind of data for the short term--be it the pseudorandom sequence or state changes or both--you may want to eventually delete it while still having the option of re-creating it at a later date. With a reproducible pseudo-random generator, you can do this by saving only the parameter settings and the seed value.