Okay I can't respond to everything you said but some really good points of consideration raised about TDP, thermals and PCI lanes. I cannot say all of those specifics were thoroughly enough looked at.
I can't figure out the overall design intention here due to the seemingly high number of design conflicts.
In case you missed the overall design intention: Shrinking the size of the Mac Pro tower box, and also reducing its total thermal output. Everything else stems from that basic goal. Meaning being pretty ruthless about whats actually required, in some cases.
Drive bays
=========
You say: "what's the harm in the 3.5" bays?" I agree with your description of their merit as a form factor for Mechanical HDDs. However its not enough to make that a pressing necessity going forwards.
How many legitimate use cases for 3.5" HDDs anymore? It depends what you are comparing to. You can buy 512GB Crucial M4s which are pretty affordable these days (vs the total system cost of Mac Pro). And the newer 7200rpm 2.5" HDDs themselves aren't really that much different anymore than any standard 3.5" HDDs performance-wise. If you previously wanted a 10,000 RPM hard drive - it makes little sense not to buy an SSD instead.
The main reason for ditching 3.5" for 2.5" bays is that the 3D volume of saved space (or conversely wasted) becomes pretty significant, when multiplied over x4 bays. I didn't calculate the savings, but if you have ever compared two real 3.5" and 2.5" side-by-side, looking at them next to each other you will understand what kind of a size difference I am talking about.
CPU
=====
That doesn't make sense with your other parameters. You want to go with mobile cpus. This means very few PCI lanes. Your typical mobile gpu actually runs on 8 instead of 16... If you're looking for a lot of bandwidth, you're actually tied to Xeons.
That's really good point which I totally forgot to look into. One of the primary differentiating factors of Xeons may indeed be its high number of PCI lanes. But what actually matters is not if the Xeons have double or more the total PCI bandwidth, but if the Mobile CPU is not too restricted and has sufficient PCI lanes to meet all of the typical demands. Biggest customer is the 1x high-end GPU, assuming you understand the SLI is a pointless waste then 2xGPU is never needed. So that saves some. But also up to 2 peripheral IO cards, and the integrated Thunderbolt, iGPU, etc. I really haven't checked but the advent of PCI 3.0 did doubling the bandwidth across the board. So maybe it is enough bandwidth, maybe not - I honestly don't know. But a in any case a good argument to make against using this particular Mobile part vs a desktop part.
You're making an entirely different computer class.
No sorry, the intention was to discuss about how to make a smaller form factor Mac Pro, and lower the TDP envelope. I think you meant to say something else comparing about the iMac and the Mac Mini. I'm not comparing any of the expectations specifically to the iMac or Mac Mini. Only to the current generation of Mac Pro. The basic premise is: what is the most allowed sacrifices can we get rid of without sacrificing the typical PC-type (ie desktop) upgrade choices and power-user options and choices over the performance that actually matters to people?
My thinking is that the one choice nearly all users do not wish to be taken away is the graphics card. So that is the one thing that cannot really be sacrificed. So must continue to allow the GPU up to the huge massive 200W of TDP headroom, ie for the fastest graphics / best performance per dollar graphics cards available. This also means you inevitably have to try to be a lot more frugal on whatever else draws power inside the box. Otherwise the overall target of reducing size and TDP is not achievable.
Your point about Desktop CPUs being only 70 Watts now - its a really good point. That is only 25 Watts more than the mobile part vs 200W GPU.
This makes very little sense. You pay more for less with cpus deemed mobile, and 45W desktop cpus exist anyway.
Indeed, Intel say that in future generations we will be given configurable TDP technology making the whole issue of distinction less problematic. However: You will always pay more for the highest binned parts with the highest TDP/watt. Whether Intel brand and sell them as a "K-series", a desktop part, or a Mobile part is really splitting hairs. The key difference was about QuickSync...
In terms of integrated graphics, you can get them on desktop gpu types as well. There isn't a reason to go this route at all if you're looking for cost effectiveness.
There is a reason, and you seem to have missed that. The key difference is that the video transcoding via QuickSync will be twice as fast. Which is actually a very significant thing (for that particular feature).
The choice of naming the quad mobile i7 was only because it is *today's* best Intel part/SKU for that. Intel's desktop CPUs may offer this aswell in the future. But until then, and for a concise shorthand, its a heck of a lot easier just to refer to "quad Mobile i7". Since then everybody knows what you are referring to. The thing is already in many other Apple products, including the well-known Retina MBP.
If anything, you can blame only the Intel policy, for crippling (halving) the intergrated graphics on all their own Desktop CPUs. Sorry for being the messenger to be pointing that niggly little problem out...
Don't forget that we are talking here about what is in Apple's control to change, and not what Intel is choosing to do. So Mobile i7 is the choice (or not) for those reasons - Since as I have already conceeded it must provide a good enough PCI bandwidth otherwise is not really acceptable either.