* Peak performance came at the number of threads == number of cores (vs hardware threads, being double that).
* How does the same hardware profile when running Linux?
Note that there are some concerns with the benchmark technique used here (both the test that's the subject of this post, and the DragonFly testing). Running the pgbench test tool on the same host that's running the database under test can severely impact the results in ways that don't reflect what you'd see in a real deployment. See this post from Attilio Rao for more detail:
The test here was done to push PostgreSQL on FreeBSD to the limit, in order to understand and then eventually fix the bottlenecks; it's not directly useful as an indication of the performance one should expect from a real deployment.