Really nice demo. Wish I could try it out (AMD here). I'm curious, are you planning on porting the code to OpenCL anytime soon or is it too deeply integrated with the CUDA platform? Also, how do the lens effects work, for instance, at the beginning of the video? Are they path-traced as well, with the optical system part of the geometry, or are they a post-processing step?
OpenCL support is highly desirable, and should be available at some point in time. That, or a generic alternative. A CPU tracer is already available (not in the demo); it produces the same output, but a lot slower (it was not built with CPU performance in mind, but with a focus on maintainability).
The post processing (hit F3 in the demo, it's off by default) is just your typical image post processing. The lensflare is pretty accurate; it's based on recent work by Prof. Eisseman of the Delft university. It does a pretty close approximation of physically correct lens flares.