The nice thing about using cameras is that there's a bunch of research already into making cheap-and-nasty cameras (with horrible optics that cause odd distortions) behave like "ideal cameras" (zero distortion).

I was thinking about using them along the lines of how the compass is used in UAVs - it's not updated very frequently (10 - 80Hz or so, compared to 400Hz+ for the main control loops and inertial measurements) and when it is updated it's not very accurate (to within a couple of degrees if there's not too much magnetic interference), but you can be fairly confident that it doesn't drift over time (unlike inertial and barometric measurements). There are a couple of filter designs that very neatly integrate this information with that from gyros and accelerometers (both very fast and very accurate in the short-term, but prone to drift in the long-term) to produce fast and accurate and drift-free outputs. In the case of the Peachy, the high-speed control is done with the drip sensor (which gives you immediate information about incremental water level changes, but may well drift over time if the drops are not quite the expected size relative to the container) and straight feed-forward on the mirrors (also prone to drift, from things like the DACs and lubricants warming up). The cameras won't immediately fix drift there, but over the course of a print they'll prevent anything drifting too far.

With regards to camera resolution: take a standard (cheap) 640x480 webcam with a 60-degree field of view horizontally. That should allow you to measure angles down to 0.1 degrees, give or take a bit. If the cameras are mounted 30cm above the printing surface then they'll be able to measure the laser dot position to an accuracy of roughly half a millimetre (if I've done the maths right). As Quertz said, the laser dot won't just be one pixel, but it wouldn't surprise me if you could consistently lock down the laser position to within a few square millimetres. No good for actually positioning the laser in realtime (especially because cheap cameras can't go anywhere near fast enough) but fine for slower control (eg. sensing that the design appears to be a little bit bigger than it should be at this stage, and therefore the output amplitude should be slightly reduced).

Using twin cameras for height estimation appears to (again, if my maths is right) give you 0.2mm height resolution, assuming cameras 30cm above the print surface, separated by 20cm. Height resolution will change over the build height, but getting it accurate to within 1mm should be straightforward. Again, it's not a perfect high-speed feedback system, but after the resin has risen a couple of millimetres you should be able to make a decent estimate of whether or not the "vertical rise per drop of water" is being calculated correctly.