I work for an ISP and we were running FreeBSD on our critical (and extremely active) recursive DNS servers and also route master machines (edge-level BGP/OSPF management routers, which are at the level right below the Cisco peer routers - those machines run Quagga). The first issues we started having had to do with random system lockups on the primary recursive DNS server (it appeared like the OS would stop the interrupt controller - pressing the power button on the machine would briefly reactivate the interrupts for about a second probably due to it trying to initiate a sleep mode). This happened on FreeBSD 7.1 and above. I went around in the FreeBSD mailing lists and saw a large discussion about it on there, with some die-hard FreeBSD fans fuming at the kernel devs since this kind of issue went completely unnoticed (was fixed in 8.0). Then I assumed that 8.0 would be fine, but we were building new route master machines with 8.0, and Quagga was having massive issues with the kernel (I tried building the latest Quagga code which didn't solve the issue) - so we had the choice of either dropping back to 7.0 for those machines or just jumping ship to Linux (Debian specifically, which is what most of our machines run anyway). We went with Linux (still have a few FreeBSD machines though), and all our problems disappeared. The machines in question were IBM x335 and x336 1U rackmount machines.
FreeBSD used to be the standard for high-performance networking systems, but they really need to get their act together and actually field-test things before deploying production code. The code isn't simply being used on some random person's toy box, it's being used in datacenters on critical infrastructure. Situations like this will make people immediately jump ship.