1. Mobile-friendly PoW (CryptoNight Lite). The current PoW is not ideal for smaller devices because the 2MB scratchpad is too large for the cache size on most mobile and lower-end desktop/laptop CPUs. A tweak to use a 1 MB scratchpad would allow it to run efficiently on lower end CPUs including some mobile processors as well as much better performance on mainstream desktops/laptops. Credit for this idea goes to the Louisd'or project (crypto_zoidberg and doe1138), although they didn't clearly explain the benefits of it.
Feels strange to me. Why sacrifice original CryptoNight?
We are going to very strong ARMs now, with lots of fast cache, look at MediaTek, look at Apple A7/A8, look at AMD servers with arms, pretty good performance for an original CryptoNight. I got your idea, but i think in that case "light" cryptonight will be less secure.
Sacrifice is a strange word to use here. I see no sacrifice.
"Less secure" how? AFAIK security of PoW depends on the amount of mining and the distribution of that mining. I'm not aware that a particular PoW function makes one more or less secure than another.
As far as scratchpad size, sure you can always find some targets where 2 MB is best (not sure about Apple though, they have 2 MB/core of L3 but only 1 MB of L2, so that is not clear) but there are also a lot where 1 MB is better, especially if you are looking at cheaper mainstream models as opposed to high end. I think there is room for both.
EDIT: I looked at AMD A1100 (1 MB shared L2, 8 MB shared L3). Although I'm not targeting servers particularly, I see no reason to think that 1 MB scratchpad is not going to be a performance improvement there while preserving the original principles of CryptoNight.