I read that as a ~2% slowdown in reference counting, or garbage collection in general.

I strongly doubt that. Adding a check to a simple increment instruction is likely slowing it down by more than 2x not 2% so it must be some whole program performance benchmark that's slowed by 2%.

The PEP says that a naïve implementation would cause a 4% performance hit and have a list of various ways of speeding up cases where immortality makes sense / is guaranteed that they say can bring this to parity. Maybe they implemented part of the list.