Quote from Reddit (03-10-2017): https://www.reddit.com/r/Aeon/comments/73tfz5/what_are_the_pros_and_cons_of_cryptonightlight_to/
What are the pros and cons of Cryptonight-light to Cryptonight?
The main advantage is better utilization on lower end CPUs that often have only 1 MB of cache per core or HW thread. This matters for mining (meaning those lower end CPUs can be competitive, while with regular Cryptonight they are not) but also matters for processing block headers/verification. Cumulatively the 2x speed improvement in iterations, a hypothetical (but common) 2x or more speed improvement in utilization, and a 2x improvement from reducing our block time to 4 minutes from 2 minute in Monero, we are looking at an 8x or more improvement in header verification.
Memory-bound algorithms with massive scratchpads (GB) can argue that they are “too big” for ASICs (or equivalently that most of the ASIC cost would be memory which is already well-optimized) but even regular Cryptonight doesn’t do that. 2 MB isn’t all that different from 1 MB and neither is that major of an obstacle. The obstacle is radically improving on the already-optimized cache memory and already-optimized ASIC-like AES and large multiply instructions in CPUs. There are several opportunities for theoretical gains but the appear vastly more limited in magnitude than something like SHA256, and still have to compete against CPUs in economies of scale and staying current with process.