How is adding these two light sources here different from (or not different from) just a single light source?

Is this parallelizable (as in tracing rays from different pixels in parallel)? If so, is there a benefit to memoizing/caching the results?

Exactly how inefficient is this computation compared to rasterization? What attempts have been made to optimize its performance?