Just finished half way through Ren Ng's thesis on Light Field Camera--Ren Ng is the CEO of Lytro.

If I understand it correctly, these are the characteristics of his LF camera:

• An array of micro lenses is fitted in front of the image sensor. A micro lens will cover a group of pixels on the sensor, so basically a micro lens will represent a pixel in the output image.
• Image resolution will be reduced. How much it is depends entirely on the number and size of the micro lenses. The minimal size of a micro lens is limited to the lens optics. Optimal resolution appears to be reduced to 1/16th (16MP to 1MP).
• The purpose of a micro lens is to capture the directions of the light. Conventionally, a pixel records photons from all directions. A micro lens would directionalize the light hitting the pixels (forcing a pixel to record photons only from one direction). One direction each, a group of pixels under the micro lens will record light from all directions. I use "directionalize" here because technically it isn't the same as "polarize".
• To digitally refocus and to get the final image, the software program simply ray traces the light, one pixel at a time. Ray tracing is normally a computationally intensive task in 3D but for LF data, this should be extremely fast since the RGB values of the pixels are already there; it's just a matter of picking the right pixels on the sensor to produce the final image. Per-color-channel ray tracing could be used to completely eliminate chromatic aberration from the image.
• Show Stopper: I'm not too sure about this but I believe such LF camera will not make the depth of field shallower beyond the lens'. The way it works, we should be seeing only LF cameras equipped with a large sensor at least APS-C size and with standard DSLR lenses. If you put the array of micro lenses on top of a P&S sensor, it wouldn't make a difference anyway since the DoF is already too deep; it would be pointless. Besides, it's probably extremely difficult to produce smaller micro lenses for P&S sensor. So my speculation is we will only seeing LF cameras in DSLR, ILC or Fixed Lens Camera instead of LF P&S.
• So yes, if you want shallow DoF, you'll need to get a lens that gives shallow DoF. 
• While can't get shallower DoF, it can however extend DoF to get everything in focus. Again, it's just a matter of picking the right pixels.
• Ren Ng talks about producing a 160MP sensor so that the final image will get around 10MP. Such LF DSLR however will greatly reduce the low light performance. It would be like losing 4 stops of light.

The promise of a LF DSLR is to allow you to refocus after the fact. Make no mistake, it can be a paradigm shift in photography. It could probably be the age of point-and-shoot APS-C Mirrorless Compacts!