Yeah the simplest, cleanest and cheapest? way to get 16GB would be 8 2GB chips. The Scarlett video I don't believe tells us anything other than confirm the GDDR6. After Richard Leadbetter stated just how much info was gleaned from the Scorpio video down to the die size, Microsoft made sure they didn't give anything away this time.
24GB/384-bit bus, 32GB or mixed density chips I just can't see happening.
Why? What is so clean about 16GB? The fact that is a power of two?
As I said, I'm a software developer(though I write web servers, not games), so maybe I have a different approach to this, but I feel this train of thought is pretty simple.
16GB of RAM likely means 256 bit bus. From what I judge from people smarter than me on beyond3d is that might not be enough bandwidth-wise for that kind of GPU. You do not build a 12TF GPU that will be memory starved, since that defies the point - you could go with cheaper GPU, smaller chip and still have the same performance. People seem convinced that you cannot get that kind of bandwidth on 256 bit bus. If you have wider bus(like 320 or 384 bit) you have to add more RAM, or have different sized dies, which doesn't seem to make sense. Ergo - it makes sense they have either 20 or 24 GB, since they need the bandwidth to feed the GPU.
It has nothing to do with SSDs and the like - they are there for faster asset streaming, not for feeding the GPU. SSDs will be huge for open worlds, but they will not help in rendering. Also, there is no such thing as too much memory.
From developer's perspective, adding RAM is never a problem, so they can change the targets whenever they want.
This is pure speculation, so take it with a huge grain of salt(and while I have better than most grasp on optimization and computer performance, I have no specialized knowledge about graphic systems, so I can easily be wrong), but I just don't see how 16GB would work with the rest of the spec.