[Bartweiss]

Yes, it can. It's fairly common to keep a pool of learners which aren't identical for exactly this reason - either playing against prior snapshots, forking training, or lightly randomizing some components of the learner.