Well, he did that to simulate a drastic flaw in a driver that caused it be highly unstable.
Some benefits he mentioned in terms of developing drivers as user programs is that they only get the permissions that are appropriate for them. Only the disk driver should have access to the disk, etc. whereas drivers that live in the kernel can get their messy hands on everything (I've had video drivers that mistakenly output sound, as well). It would also help with driver development, because if the driver dies, the OS remains stable (rather than needing to reload the OS). There are only three or so absolutely essential processes, and he emphasized that they must be kept small to reduce the likelihood/sheer number of bugs.
I found it really fascinating. IMO, if MS knows what's good for it, they should be prototyping their 5 or 10 year out OS on this basis. Windows has just been ballooning in size, such that it isn't really sustainable.