Cost of die area is not a linear function. There are "sweet spots". That's why PS4 die area ~ xbo.
That's why PS4k exists. It does not cost much more than base PS4.
Cost per gate graphs indicate that the PS4 die will indeed cost rather more than the APU in the base PS4, and "sweet spot" is affected by how the die is used (i.e. sram is defect tolerant) and the price the market is willing to pay for the chip (both MS and Sony wanted to come in at a similar BOM). I think you are rather oversimplifying the matter. But that is not the point I'm making.
The chip is only a fraction of the price of the system. The PS4 teardown from launch suggested that the APU was $100, with a BOM of $372. If Sony had paid twice as much for the APU, the console would not have been twice as expensive, it would have been ~25% more expensive. But that chip would have been vastly more expensive to support (memory, power etc) so it could never have been realistic.
"Sweet spot" of a chip for a console has far more to it than yield vs die size.
With the perf/watt and perf/GB/s advances of 14nm and AMDs new architecture, Sony can now house a significantly faster chip without incurring outlandish extra costs elsewhere. HDD, BR, case, cooler, power, even memory will all come in somewhere close.
That is why Sony can launch a significantly faster system - with relatively little extra cost - and that is why Neo exists.
Edit: and as GDDR5X / DDR4 / LPDD4 / HBM2 mature and as yields on 14nm improve, and as architectures are refined, and as markets for higher end systems rise or fall, "sweet spots" will continue to move.
