There are basically three ways to do RAID: RAID 0, RAID 1, and RAID 5 (RAID 2-4 were inferior examples used in the original RAID paper to gradually introduce the idea of RAID 5).
RAID 0, or "striping" (one 'p'), can be done with any number of disks, and the total storage capacity is equal to the combined storage capacity of all the disks. When data is written to disk, it is split across each of the disks in the array. This provides a performance benefit when reading and writing, as the operations can be performed in parallel on each disk. Since hard disks are so slow (in relative terms) these days, this can be beneficial.
However, if ANY ONE of the disks in your RAID 0 array fails, you lose all your data. Which means that a 2-disk RAID 0 array is approximately half as reliable as a single disk, a 3-disk array about one-third as reliable, and so forth.
In general, it's a terrible idea unless you really need the performance and have a good backup strategy.
RAID 1, or mirroring, can also be done with two or more drives. The idea here is that each disk in the array is an identical copy of the others. If one disk fails, you don't lose your data. Read performance can be increased, because you can read from each disk separately, but write performance is unimproved. Reliability goes up exponentially with the number of disks in the array. However, your storage capacity is only equal to that of the largest disk.
In general, most people are happy with using RAID 1 with two disks - a third is probably unnecessary.
RAID 5 can be used with three or more disks. The total storage capacity of the entire array is equal to that of all the disks, minus the capacity of one of the disks (in your case, 1500 GB). It can survive the failure of any one disk in the array. Read performance is comparable to RAID 0, write performance is comparable to RAID 1. The advantage is storage capacity.
The way RAID 5 works is somewhat complicated. It uses a technique called parity. If you're curious, read on.
Basically, assume we have three disks: A, B and C. We assign A and B to store data, and C to store parity. Every time we write a byte B1 to a certain block N on disk A, we read the contents B2 of block N on disk B, and store the sum B1 + B2 in block N on disk C.
Now, if we lose disk A, we can reconstruct the contents of block N on A by reading block N on disk C, and subtracting block N on disk B (giving us (B1 + B2) - B2 = B1). Likewise for disk B. If the parity disk fails, we just compute all the sums again.
This is basically how RAID-5 works, except that the parity information is not stored on a single disk (for performance reasons), but distributed among all the disks in the array.
If you have three disks, RAID-5 is your best shot, since you'll get the most storage capacity while maintaining redundancy. Read performance will be approximately equivalent to RAID-0, and write performance will be approximately the same as a single disk.
However, you will need to buy a hardware RAID card, as I don't think Mac OS X supports software RAID-5.
You can also combine RAID levels pretty much arbitrarily - RAID 5+0 is a RAID 5 array running on virtual disks that are themselves RAID 0, for instance.
