That annotation is due to disc handling changes in OSX.
I can't say with certainty, but if you are running OS 10.6.1 or later, drives larger than 1TB are supported. I think you may be safe to use RE4, but you may want to order only one drive to start, and ensure you can use them.
If not, the RE2 will work under OSX 10.6.1 or higher, and so will the Toshiba drives.
What I suggest is that you use a pair of 2TB drives as your base array inside the system, hosting your OS, and user accounts, and install an SAS or eSATA card, to use an external enclosure for larger drives. I assume you want to do RAID-1 for the two drives?
The problem you will face is that internal drives, under a RAID-1, are software-controlled. If you want good RAID support, and options, you want a RAID host adapter, and an enclosure.
An SAS adapter gives you the most options, as you can use SATA HDDs on it, as well as SAS drives, and accessories such as tape drives.
I further suggest using the third internal bay as a backup mechanism. Every so often, clone your two-drive RAID-1 array onto that drive (drive three), so that if anything ever damages the RAID-1 array, you can use the third drive to restore it, or to boot in an emergency.
Really, three 1TB drives would be your best bet, and then use an external enclosure to make a RAID 5, RAID 10, or RAID 51 array, on which you can host your media, user accounts, and databases. This gives you massive storage potential, data security--redundancy and reliability based security--(RAID mirrors are the best for this, but the most expensive), and speed.
In terms of speed and data redundancy, here is the simple breakdown:
RAID 1: Two drives, mirrored. You lose 50% total capacity to make the mirror. Read speed is faster than JBOD, but write speed is diminished.
RAID 0: A stript of two or more drives: This gives you greater speed, but is very easy to break, losing all datum. Read and write speeds increase compared to JBOD.
Raid 10 (1+0): Combines RAID 1 and RAID 0 to make a stripe of mirrors. This gives you improved read speed over JBOD, and improved write speed over RAID-1 alone. You lose 50% capacity in the mirror, so eight 2TB drives give you 8TB of base space, minus RAID-0 allocation and filesystem space, for around 7TB total on 16TB of disc space.
RAID 5: This uses parity for data protection, while striping drives, and is not as secure as RAID-1. It uses less space (33% on a three-drive RAID-5, 25% on a four-drive RAID-5, and 20% on a five-drive RAID-5), and has moderate read and write.
RAID 51: A RAID level that is the most expensive to maintain. It takes RAID 5 + RAID 1, to make a mirror of RAID-5 sets. The mirror is dead space, so in a six-drive RAID 51, you are using 33% of each mirror for parity, and of the remaining space, you are using 50% for the mirror. In a ten-drive, you are using 20% for parity, and 50% for the mirror. Thus, a ten-drive array of 2TB drives (20TB disc capacity) uses 4TB for parity, leaving 16TB, then uses 50% for mirroring, leaving 8TB total out of 20TB. It is useful for situations where you need the highest level of datum redundancy and read/write speed.
Finally, you have RAID-50, which combines striping and parity of RAID-5 and RAID-0. This is somewhere between RAID-5 and RAID-10 in terms of speed and data protection. You sacrifice the full mirror protection of RAID-10, for more space, and have better R/W speed than RAID-10. For example, eight 3TB drives as RAID-50 would give 18TB usable space,out of 24TB disc space.
If you use the internal bays, your only options are RAID-0, and RAID-1. Of these,RAID-1 is what you want if you desire a self-sustaining mirror. Apple Disk Utility supports software RAID at level-0 and level-1: Keep in mind that this uses CPU cycles, as the XServe does not have an internal RAID controller.
You, as an alternative, could use one 2TB mech, and two 1TB mechs. Set the two 1TB drives as RAID-0, for speed, and do manual clones on a regular (e.g. weekly) basis onto the 2TB single drive.
If you put your user accounts on an external array (useful), be sure to keep at least one admin and one user-only account on the system boot drive,in case you ever lose the array. My suggestion here is to keep your main accounts on the internal drives (with clones of them on an external array),and all other users on an external array.
This will ensure you never have access problems, and will both give you more space on the internal drives, for SQL and applications, while placing other users on the external array, for faster disc usage (not needing as many r/w operations on the system drive also increases its lifespan), and allows you to keep user accounts isolated from the root filesystem.
Anyhow, again answering your question, I believe that 4TB drives should work internally, but it is based on the OS version.
Apple don't list larger drives in that data sheet because they never sold any Apple Drive Modules in capacities greater than 2TB, and Apple do not officially support any hard drives that they did not sell directly.
I use WE2 and WE3 drives in my array,and Samsung 1TB (enterprise) drives inside the XServe itself.
