I can see a few reasons:
1. He just wants to manage one VMFS volume rather than 2-3.
2. When GPT partitions come out (that is if :)) you can migrate the same file structure to the new VMFS without having to change anything.
Since I was suggesting concatting LUNS within a single drive, the potential loss of an extent would generally fall in the failure domain of the drive, so this doesn't really add any substantial risk. As of vSphere only the head extent needs to be online, and I have many customers that use extents and have never lost one. I am not saying this is not a possibility, however the failure domain has been reduced in 4. Since the metadata is captive, recovering bouts of online/offline is reliable within the scope of a single extent VMFS volume.
In most arrays now, LUNs being presented are virtual not physical RAID volumes, so loosing a virtual LUN would probably be tragic using extents or not.
If you have an array now using vStorage (VAAI), ATS is used rather than SCSI2 so locking events go down dramatically and scalability using extents reduces the VM/LUN issue that is a real limiter these days.