Forward-compatibility with distros is always a factor. In the past I've always brought up a new box on new disks, then added in the old disks to migrate data across. With only RAID1 this works well. Obviously this gets harder with RAID5 because you need all the disks in the array (you can't just mount a degraded half like you can with RAID1), and the number of disks is higher. I'm hoping that the recent mdadm tools and LVM will enable me to grow the RAID5 array as necessary in the future. That all being said, I don't see XFS, JFS or Reiser being dropped from mainline distros anytime soon and I'm only going to run mainline distros on a fileserver - I'll leave the esoterics for the desktop.

I'm still not sure how much I care about performance versus convenience and reliability. For the RAID5 array, read performance is a bigger issue than write - mostly the data is static. Grabbing shows off of the replaytvs is severely bandwidth limited by the replaytvs themselves. Writing mp3s is going to be CPU limited. And it's not as if this box is going to be serving many people either.

The smaller RAID1 system array is going to need better write performance. syslog.

A month or two ago I saw an in-depth benchmarking of all of these filesystems, specifically with RAID. I wish I could find it now.