More cores _actually_ running simultaneously means more likely chance of encountering sporadic race conditions. Also makes it easier to measure (and thus work to improve) high thread-count scaling.