The fix we put in to make pw's writes safe (fix for passwd file corruption) also made it slow in some circumstances, especially with large numbers of users. Short of thousands of users, I haven't heard of any delays of minutes attributable to that. In FreeBSD 10.3, a different fix for that problem has been implemented which doesn't have the performance issues in those circumstances. I've put it through our power cycle test rig upwards of 3000 power cycles immediately after passwd write, and it still survived fine. I haven't tested large scale performance, but the FreeBSD developers who reviewed and implemented the change have. So any portion of it attributable to that will be significantly faster in 2.3.