RAID 0 is the misnomer of the decade. Don't do it, as it doubles the chances for data loss. The "R" in RAID is supposed to stand for redundant, but there is none in RAID 0. RAID 0, AKA data striping, attempts to improve performance by splitting the data (which I'll represent as ABCDEFGH) across two drives, like this:
Drive 1:ACEG
Drive 2:BDFH
There are two problems: First, if one drive fails, the data on both drives is lost. Second, while striping theoretically can double the performance of a single drive, with IDE (yes, SATA is just a new flavor of the same old IDE), it doesn't actually work so well in practice because IDE does not perform particularly well at multiple simultaneous access. It does improve indicated performance in benchmarks because IDE striping can improve sustained transfer speeds, but everyday use is generally not about sustained transfer; it's about reading and writing small chunks of data at a time. (Striping across two SCSI drives would be a different story, performance-wise.)
Redundancy is only acheived by mirroring the data, like this (RAID 1):
Drive 1:ABCDEFGH
Drive 2:ABCDEFGH
Or like this, by combining the two techniques (RAID 0+1):
Drive 1:ACEG
Drive 2:BDFH
Drive 3:ACEG
Drive 4:BDFH
For more detail, you might want to give
this AT article a read. They evaluate RAID 0 vs. single drive performance using 10,000 RPM Western Digital Raptors, but the same issues apply to any IDE RAID 0 setup.