This Firewire-800 external drive enclosure from DAT Optic, model qBOX-F looks kind of tempting. The bare enclosure is $199.00. Internal interface is IDE unfortunately, although the asterisk-footnote says you can use SATA if you use a IDE-SATA bridge.
DAT Optic qBox-F drive enclosure
So the next question is, how much of a performance penalty for using the bridge? This article from May 2003 from TechRepublic states:
The problem with bridge controllers is that they eat up bandwidth, and quite a lot of it: on a 150 mbps-rated device, a bridge controller may leave you with only 70-80 mbps bandwidth.
Granted this article is from 2003, so maybe a lot has changed in 4-5 years? No idea. But 50% bandwidth penalty seems pretty harsh.
Okay, so let's ignore that for now and see what it would take to use this for our home ZFS solution.
If we populate with 4x320GB drives, at RAID 1+0 that'd give 640GB (I know, that's marketing GBs at 1000Mb/GB, not real GBs) or 960GB with RAID 5.
Today's Newegg.com has Seagate Barracuda 7200rpm 320GB SATA OEM drives for $80 apiece. So figure $320 for drives, plus $11 per drive for IDE-SATA bridges. I suppose we could opt for the equivalent drives in the IDE flavor, but IDE wouldn't carry forward too well into other computers, if/when they get traded out for newer drives. The same 320GB drives in IDE cost $85, so for just $6 more apiece, I can do SATA + bridge.
This adds up to $544 for enclosure + drives (not counting any shipping costs). Not bad. If my main system can be the ZFS host then I'm pretty much done. Otherwise to get host system for around $500 to stay near the $1000 limit would mean a Mac Mini or maybe a small mini-ATX x86 setup. I'm not as well-versed in all the PC components and motherboards and such, but I'm guessing $500 would be possible for a simple Solaris 10 x86 system to handle ZFS? Or correct me if I'm wrong.
So this is sort of where I'm aiming at, but the unknown factor for this enclosure is the IDE-SATA bridge performance penalty. This web page casually states "expect a slight (~10%) performance penalty", but there's no support for this claim.