this all comes down to time. i can reapply the data to a fresh VM image in a matter of hours and have it back up and running, pretty much without variation. hunting down a deep, dark problem can take 30 minutes or it might take days, and depending on the problem, that may simply be unacceptable.
the real skill is knowing when to pull the trigger on a rebuild vs knowing when it's something you can find and fix. hunting down problems and fixing them is something many sysadmins crave. at least VM's give us the ability to investigate the broken machine at our leisure, while a working VM can jump into production.
unless management wants to rely solely on rebuilds and the time investment it takes to do them every time, there will always be a need for sysadmins to analyze problems and figure out the "whys" and "hows" that caused them.