So Spec, I'm certain with regard to (1) that there are multiple back-end lines to one substrate. If we return to "Herbie" for a second, if there weren't, the throughput limit would be some post-cutting operation automatically which could literally only handle one unit at a time. And that would be insane given that you could be making 6 or 9 or 12 panels at once at litho/masking/ITO patterning/whatever.
So I don't know how many lines there are after cutting, but it's going to be several and since it's not going to be 9 or 12 in all likelihood, you are almost certainly correct that's "Herbie". So having established that, we can do a little thought experiment that I will admit is fake, but the math works out, so bear with me.
So the Sharp plant can make a 6 x 70" motherglass or it can make a 2 x 70" motherglass and then 16 x 35" panels. In the first scenario, it could split the cut panels down the 2 lines in the plant that handle 70s and process 2 panels at a time for LC filling and sealing. Let's just pretend one of those in Herbie and the plant has 4 filling/sealing lines but only 2 of them can handle the 70s at this point. In the second scenario, the 70s are shunted to just one line and 3 lines are using to handle the 35" panels. Each has to handle 12. The good news is they can handle them much faster on average than the one line handling 70s. The bad news, the 35s filling/sealing are now Herbie.
From a pure cost accounting standpoint, the cost of 70s has been driven up and they are suddenly looking less good, but if you need to make that many 35s, it just doesn't matter. You had another choice which was to just run 70s on big sheets all day and then shut down everything and just run 35s the next day. Maybe they do it that way, but that's costly too. Nothing is free.
Honestly, I don't know what they are doing. But so long as Herbie is post cutting, the notion that doing same size panels is dramatically more efficient is false. Everything that happens on the non-Herbie part of the production system doesn't need to be optimized time wise and -- this is the hard part to wrap one's head around -- the system gains nothing from optimizing it time wise, in fact it might lose something.
If you get too good at making patterned cut glass but can't fill/seal it quickly enough, you build up too much WIP along the way. WIP is not good. It's messy, it requires special handling in this case. It can break. Ideally, you want the faster process to be just a bit faster than the slower/bottleneck/Herbie process.
Just to clarify (2) above, I understand obviously that a specific fab handles one size of motherglass. What I'm guessing is that the machines that do the pre-cutting steps are highly computerized and able to do their work regardless of whether a sheet is all big panels, all small panels, or mixed panels. And quite frankly, I doubt the efficiency losses of mixed panels are more than a few percent in time (which we've shown does not matter) and perhaps a few percent in materials (which could matter, but it should be fairly small).
The likelihood is that the post-cutting machines are more flexible rather than less flexible, but that a given line is ideally running a given size at a given time. There are things about handling and testing that are going to be easier when that's true. But since we've also shown that are almost certainly multiple lines post cutting, it's again not especially clear that having all of the same size is especially important or especially efficient.
Now all that said, there are steps in manufacturing whose current state I can't begin to understand. For example, the two sheets of glass are separated by a series of spacers that require a fairly even distribution. Whether that step is trivial for multiple-sizes of display on a single sheet of motherglass, I really can't say. I can say that screen printing or flexo printing is absolutely trivial for such a purpose. Spacer distribution? No idea.
Now, as for the question about depreciation, if Sharp took a markdown on the plant in the U.S. before it even produced anything, that'd look pretty bad. But Japan has its own set of rules. To be completely honest, I would not spend a lot of my time worried about per-plant depreciation if I were CFO. I would worry more about a reasonably proportional depreciation allowance per product I produced. That would make newer products in expensive new plants automatically look better. Financials reported out to the public, however, don't contain this data anyway. They contain consolidated financial data.
In terms of pricing, there's a legitimate question about what product lines to "bill" for the new plant. But again, I'd probably bill my entire LCD operation. That makes the 70s look cheaper to build even if it means my cost to produce 32s in my old plant has gone up by $2-5 per unit. I don't personally think that's bad internal accounting. It allows me to be always forward looking and aggressive. If I keep charging my new product the full freight of new factories, it always looks expensive vs. existing product. Bad way to see the world and not an entirely accurate one.
Compare Sharp to a company that builds nothing. That company will pay whatever their contract mfr. charges and that contract mfr. will use as state of the art a plant as is needed to build the item. Apple might need Foxconn's newest factory for iPad2, but an unimpressive 2006 one for Mac Mini. Foxconn is going to charge Apple more proportionally to assemble iPads because they are more complex, but not because the plant is newer. That's nuts, especially because Foxconn might well use the newest factory for Mac Mini as well (for whatever internal reason).
If Sharp simply takes all it's LCD fab depreciation and "bills" each panel it makes a proportional share based on size -- regardless of which fab it's produced in -- it's going to make the 70s look a bit cheaper than they would be for a startup with the exact same cost of an LCD fab as Sharp. But that doesn't make Sharp's accounting particularly questionable.
There's a saying about "everything in moderation". If only it was applied to well, you know...