Help with branched path tracing!

So i was able to cut the render times of some tests nearly in half and I’d like to know why as well as what you think would be good enough for some of the samples, but here is my guess…

Diffuse does not take many samples right? Even in a complex scene? Can i keep this low, like 256 tops?
Glossy take more obviously.

But the others like AO, transmission (what do these do?), and subsurface, say I don’t have anything that needs those samples, will i benefit by leaving them at 1?

Also, why is it all multiplied by the AA setting?

John if I’m not mistaken many animators are not using branched path tracing. With that thought in mind set Diffuse to one (1) Since Diffuse will run your render times up more then anything in that table.

Why do you think that is?

As far as I know AA is basically a whole sample for the pixel among other pixels (around it), while the rest work to create that one pixel by reading light bounces. I have no idea what antialiasing has to do with it though, that was just my take on it.

What I would like to know is whether anyone has experienced crashes with branched path tracing. So far it crashed my render every single time I tried to use it (on 2.75 and former) while not-branched PT worked fine.

I think the why-not-animation question is interesting as well - the way I see it it is perfect for animation since it can cut rendertimes immensley. Then again, perhaps it has to do with the scene changing in animation so while you may not have many volumetric materials in one frame you may have nothing but that in about 5.