I think that WeBWorK problems should show consistent output when loaded with the same seed. Since there are mechanisms for re-randomizing problems within a homework set, I don't think it is necessary to perform the re-randomization within the problem. Also, for me the expected behaviour is that when I load a WeBWorK problem it will always show the same version unless I've specifically requested that it be re-randomized.
This is the way that PG behaves, so it only gets complicated when using external integrations. The R server connection has been set up so that the problem seed from WeBWorK is passed to R, which means that the behaviour is consistent with PG. I would suggest that for other integrations (Sage, GeoGebra, etc.) that either the randomization be done in PG then passed to the external system, or that the problem seed be used to initialize the environment in the external system.
Not sure if this warrants its own thread?