If I may try to put this into layman's terms (I'm not sure if you were being sincere with this post or snarky)... this refers to lighting effects particularly geared toward shading, anti-jagginess, and the color palette? Am I hitting the right ballpark?
Well, UE3 pre-DX11 is essentially just a forward renderer, so the memory reqs per render pass are significantly lower (depth + single frame) whereas the full deferred shading buffers are at least 4 render targets + depth. On console, they typically don't use FP16 buffers either due to memory and the rate at which hardware processes said buffers (it's half-rate for FP16 vs RGBA8). (Depth remains 32-bit per pixel)
Samaritan was put together with the use of 4xMSAA in mind, so that quadruples the amount of per pixel data. Again, on console, we typically see a lack of MSAA (2x was used in earlier UE3 titles, but only for certain render passes, usually skipping it for post-processing and lighting/shading).
So... "6 years" of progress when comparing Samaritan to your typical UE3 console title has gone into three extra render targets, double the memory per buffer with a higher precision format, 4xMSAA for every buffer (including depth, and even the forward rendered passes for transparencies*), and several times the number of pixels. It's a big jump in memory and bandwidth reqs.
*Transparencies e.g. hair was supersampled with the 4xAA.
-------
Ultimately, Samaritan is just a tech demo with ludicrous settings.
