Linux SW Raid: whole disk or per-partition?

I have inherited a machine which has 2 physical disks and uses Linux SW RAID(1). Both disks are partitioned and are are all individual arrays (/dev/md0, /dev/md6, etc.). Those arrays are then mounted (/boot, /home, etc. even /tmp).

As RAID is designed to mitigate physical failures, is there any reason why one would use this technique over whole-disk arrays that are then partitioned (perhaps using LVM)? This seems prone to more potential issues, but may have some special properties that I haven't been able to glean.

I'm planning on moving this setup to: disks→SWRAID(1)→LVM as I'll be making multiple VMs out of the one machine, but wanted to make sure I knew what I was doing when I got rid of the old setup.

Unless I have some requirements involving putting different data on different spindles (which you can't really do anyway with two drives in RAID-1), I prefer to use a large partition to allow the most flexibility with LVM. It is certainly much easier to resize logical volumes than it is software raid volumes.

Just remember to install grub on both drives.

Setting up separate raid devices for each filesystem is needlessly complex. Your idea of one big raid device with lvm on top of that is definitely the way to go.