Based on my limited experience, some potential issues:
Not having the user's fingers obscuring the view of the screen may in fact increase the perceived latency, since they're focused on the screen and don't have the motion of their fingers to distract them.
The latency issue would be twice as bad if you want to render a 'ghost' of your hands on the screen as described in this design concept.
Interacting with onscreen elements is more difficult when using your hands on the rear of a device, even with a 1-1 mapping. I don't really know why this is, but even with a cursor onscreen, I have found it to be true.
Accidental interaction is 5 times worse with a rear touch panel. Apps on the Vita that use it extensively are a huge pain in terms of accidental swipes and touches, especially if you try to lay the device down on a surface for a moment, or set it on your knee to use the front touch panel.
The core problem with virtual buttons/joysticks/gamepads is that you have no physical feedback about where your fingers are, and as a result you lose your 'centering' and your inputs end up being misinterpreted or not landing. Moving your fingers to the rear of the device makes this worse, because you can no longer look at your fingers to figure out where they are.