Oh, see I was thinking something more like this:
I'll be the first to admit that things got a bit goofy, but I basically had two goals: 'optimal flow' and self similarity. These mean different things, but essentially imply the same design. (For simplicity, optimal flow = if a belt is not compressed, it is a yellow belt)
So each stack fills a yellow, and the other tier belts are where things join up. Problem is the ratio between belts. Two yellows merge nicely into a red, but two reds gives one blue and one yellow. That's why I said it wouldn't be just one blue, because that's only three yellows, and then the red belt just feels useless. (plus even numbers/symmetry are my friends)
For just a single blue belt I'd probably do something like this:
And having now laid that out it makes me sad how much more efficient it is. Because
that's just not how things flow in reality I let my love of powers of two get in the way of my design goals. Especially since I can even overlay the ore input belts since each stack only pulls half. Ah well, so much for branching structure. Not to sidetrack the thread, but is there any particular reason for the way belts scale? Seems like it should be 1x/2x/4x or something, not 1x/2x/3x. Couldn't find any discussions from googling.
And just for the sake of max design discussion, I don't think your version of my layout would work (though I didn't test).
12 smelters seem to perfectly compress a yellow belt, you have... 13 and 22 on each side? 12 compress one lane (I'm dumb) so this should be fine. Looks like only one stack uses underground belts to compress. And it'll bottleneck at the two red merging into the blue, though I guess that would cause the blue to compress.
I was going for more nesting, rather than stacking in a long row. As I said I'm hoping to scale much larger than one or two blue belts, this was a first attempt. I think there may be a break even point as far as space usage. Just because this would (in theory, I haven't figured out how to route things through yet) nest into a big square rather than a row of columns. But blue belts may be too slow for that to work out, the extra rows of belts hurts compactness.
Anyway, thanks for the feedback! Do you have any thoughts on efficiency modules, other than a preference against them? Ideally I'd like to get as close to -80% energy consumption as possible by balancing them with other modules/beacons. But obviously I still need to work on my layout, and then that'll change the compression of my yellow belt too.
EDIT: I'm so stupid -- I think I confused myself by working up from compressed yellow, rather than down from compressed blue. I got to the double red belt and got confused because I already had more throughput than a single blue. Regardless, here's my submission for maxed design:
Layout isn't optimized, but I don't think there's too much to change. Please ignore missing modules; I didn't have any with me, and I wanted to post this because I realized I was totally thinking the wrong way. I can still do branching, but it'll be once I get multiple blue belts. I think I just really wanted belt throughput to be exponentially scaling (I still do, but I used to too). Red belt still feels useless, but that's just how it works because it's linear scaling. So it goes.
EDIT2: aaaaand when I tabbed back it was broken because the splitters switched sides. I thought that might happen. Just need to side load the belts though and it'll be fine. Was planning to change that anyway.
. But the crude oil is unlimited source and if you stuck on many 0.1 pumjacks you dont have to deal this problem anymore, just put a speed modules into pumjacks. There are obviously many places that are fit for productivity modules (a rocket silo for example) but there must be some parts that are fit for the efficiency way.
is my thought for a smelting design (healthy criticism welcome; I ask you do better than 'those modules suck' though). Not compacted yet (or measured), but the bigger problem than the smelting layout itself is bringing in enough ore without a 36 lane bus. Still working on how to do the drop off/pick up though. 
