He may be correct, but I would agree with you that you need a minimum of 2 DC's for redundancy...
God forbid he has a problem with the DC, and the backup....
If you lose the forest root DC you can seize the FSMO roles, and as long as you either do a metedata cleanup or never restore the forest root DC this isn't so much of a problem.
While the forest root DC is still available, I would transfer the FSMO roles onto different DC's within the forest, so that when or if the server that hosts a FSMO dies only one of the roles is affected.
I've seen a situation a few times, never been able to correctly identify the cause, where when the FRDC is offline, nobody can log in...
I would like to see a test where he turns of the FRDC, and proves that it won't affect the other domains, and then I'd leave him to it.
People, Process, Technology. As long as the IT team know and understand when and how to restore from a backup tape, if this is their DRP plan, then in some respects it's as good as any.
As regards restoration, the SP level, HAL, (number of CPU's, IDE SATA, SAS or SCI etc) need to be the same or the syste state restore won't work. So the servers in question should be pretty standard.
I would prefer to see a DRP test if it were my customer, so that we know in the event of a failure exactly what to do and exactly how long it will take to rebuild
"The isolated knowledge obtained by a group of specialists in a narrow field has in itself no value whatsoever" : Erwin Schrodinger