Is swap itself a file system at all?

Normally, no.

The exception would be if you were using a regular file for swap. (Uncommon.)

Treat swap as a raw partition or format.

Though your answer is correct as far as Linux and freebsd goes, it is INCORRECT when examining Solaris. Under Solaris 8, and I assume later revisions also, /tmp is also used as swap space. There is no separate /swap partition. We've been burned more than once at work with people putting large files in /tmp for 'temporary' storage, and having our large E15K slow down significantly.

I've never quite understood the reasoning behind Sun's choice.

I would assume that the kernel is doing low-level writes to swap--using the equivalent of dd. Thus it only needs to know what zone on the disk it is allowed to use.

Well, given that this is "Linux - General" who cares about Solaris ... ???
All jokes aside, that does sound *interesting*. I did install Solaris 10 when it first went public, but it didn't appeal, so I haven't bothered looking at it since.Where's jilliagre when needed ???.

I'm here
Nope, swap partitions are very similars whatever the Unix variant.
With Linux, while not a real filesystem the swap partition must have a structure which is created with the "mkswap" command. Solaris and probably *BSD have no such requirement.
That's the way around. It isn't /tmp that is used as swap space, but swap space that may be used to store temporary files. This is happening only if no more RAM is available.
It's curious to hear of such a poor configuration on that class of server. /tmp VM usage can simply be capped to avoid RAM exhaustion with the size=xxx mount option.
The reasoning is to have very fast read/write performance for temporary files, by storing them in virtual memory, which is RAM in nominal case, and not on disk.
By the way, this Solaris feature has been implemented later by linux http://wiki.linuxquestions.org/wiki/Tmpfs , so it mustn't be that bad

Allow me to disagree - recent thread here.
It may be proposed regularly, but I don't like the concept.

About what ?
Sure, a machine with so little RAM available will not take advantage of tmpfs. Just the opposite actually.
tmpfs is a feature you can use on linux or Solaris, but it is not mandatory in either. It can easily be disabled on Solaris if RAM is sparse.
The concept is to enhance performance, and it saves energy.
Why don't you like it ?

Polluting the page cache cannot be a good thing.
Linux has aweful (non-existant) performance/tuning metrics. The effects can't be measured. Hence I don't like the concept of using (real) storage for temporary file systems.

Every so often I get the feeling that things like Dtrace might just convince me to go back and look at Solaris again.

I'm missing how application temporary data relates to pollution.
Then you support tmpfs, which does its best to avoid using real storage for temporary files, or am I missing something ?
Add ZFS and zones to the list

PLease answer my question exactly!
I expected that you focus only on it
Is swap a File System eventually or not?

You will not get much help here with this kind of belligerent behavior.
Suggestions:

1. Read what has already been posted and ask questions
2. Say "please" and "thank you"

Hasn't it been answered ?
You expected wrong. Your question was less interesting (at least to me) than the debate it triggered.
No, it has never been, and likely never will.
Swap is meant to store pages of memory that can't stay in RAM for some reason. Pages aren't files. Using a filesystem to store them would be overkill.

I thought FSoftware Lover was making a bit of a joke with that last post, and to be fair to him/her this thread does seem to have wandered a little off topic. FSoftware Lover, why do you ask? As someone already mentioned, as far as I am aware a Linux swap partition has some kind of structure, and in that sense is a filesystem. By contrast, an OpenBSD (and presumably some other BSDs) swap partition is just raw data.