If testing a software system under the most demanding of situations isn't the way to test a practical system's performance/reliability then let's just all go with MINIX, or a Haskell-based (mathematically supported) microkernel...
If FreeBSD were pathetic, Apple, Juniper, Netflix, etc, would have failed to build they're products with it's help (think that, although the primary reason for FreeBSD could have been licensing, would they be equally succesfull if chosen MINIX instead? After all, MINIX is BSD licensed...)
- lacks support for in-tree ZFS or something comparable (btrfs is not comparable)
- merges buggy code and then tags releases before the code is production quality.
- is limited to GCC and often relies on quirks in the GCC compiler.
- does not have anything equivalent to DTrace (systemtap is not there yet)
- lacks dump devices that could helps people can debug kernel panics.
- lacks its own userland and often uses a userland developed for the GNU operating system. This causes integration problems between the kernel and userspace.
- has a known deadlock in the VM subsystem that will cripple the system when triggered. Getting it fixed is a pain because certain kernel developers are more interested in patching that affect select groups of people than they are in fixing bugs. One prominent Linux kernel developer actually made clear to me in an email that he believes that any kernel bugs that do not affect things he considers to be worthwhile in some obvious way should be left unfixed, even if patches are available.
- only works toward POSIX compliance (or anything else for that matter) when it could help software run on Linux (but not when it could help software that runs on Linux run elsewhere).
- lacks decent in-tree OS-level virtualization. The out-of-tree solutions are not quite as well implemented as FreeBSD jails, which are far superior to the Linux concept of a jail.
FreeBSD is superior in all of these areas. Some might disagree, but there is a difference between implementing something well and implementing something in a way that lets you claim to have a feature, even though your implementation is inferior.
I understood your Microsoft and Solaris argument, and again, I agree; it is possible to argue that MINIX is more reliable than Linux or FreeBSD, and that doesn't imply it is going to be seriously considered. I just think netcraft's argument is overall valid, and calling FreeBSD pathetic is exagerated. But again, is your opinion and I respect it.