ReiserFS is weird. The guy who runs that project beleives that each Filesystem should be completely constructed from scratch each time, instead of the evolution that most software needs. Gets good results, the next reiserFS is suppose to be something wonderfull. So each reiserFS numbered releases are totally new.
XFS is the best right now, IMO. Plus they just incorporated it's support in the 2.4 mainstream kernel and will probably show up in the next 2.4.x release, if I am correct.
XFS's most strong point is how it handles very large files. Like +3 gig files. Makes it good for databases and such. SGI created it as a Unix filesystem to for it's multimedia computers. Very wonderfull, giving it to the linux community was one of the better things that ever happened. I call for a blessing on SGI.
Right now the limit for it's size is 2TB filesystems on a 2.4 series kernel and a 32 bit machine and the practical limit on file sizes is like 4GB, but with a real limit of something like 16TB, or something which doesn't make sense to me.
In a 64 bit computer and kernel 2.6 you should be able to handle file sizes of 64TB! with a filesystem max size of like 4 exobytes. Maybe more, XFS is designed from the ground up as a 64bit Unix filesystem, with POSIX-combatable ACLS and everything.
XFS + 2.6 (when it stablises) + 64bit machine = one of the best massive file/database server OS in the world.
Otherwise ext3 is a actually pretty good OS and is very stable. Nice to be backward compatable with ext2 which makes file recovery easier on a damaged machine or hosed OS. It's what I use, because I've been playing with different kernel patches and different features that are incompatable with the xfs patches, but now 2.4 will have XFS support standard and 2.6 will have it by default any future machines will probably be XFS.
Or maybe JFS, which is suppose to be nice to. Of course if the SCO lawsuites turn out badly for IBM, stuff like jfs support will end up having to be pulled from Linux.... JUST KIDDING.
SCO: IBM + Linux OWNZ JOO.
(BTW a judge has ordered SCO to produce the specific stolen code or they will otherwise probably have the court case dismissed.
here for some of the details) Also SGI got in trouble with SCO a bit over the XFS stuff, so this court case affects that too a little bit.