Where did you get the info that S3 production cost is 0.4$/GH? Hopefully the production cost is cheaper for S4.
It might be a little lower as the chips are overclocked and no PSU is included, but not by much.
It is accurate for S4 though (more chips, included PSU...).
People often don´t realize that there is a real limit to the amount a vendor can discount its hardware, and that is production cost.
With these low bitcoin prices, high difficulty rates etc. the profit margins on the hardware side aren´t what they used to be.
Chip cost, cost of assembly, pcb and PSU cost are all factors that cannot really be influenced by the vendor (other than bulk discounts, these do not apply to chips though)
A good chip design can decrease the cost per GH for chips, but once everyone is on 28nm, there is no big difference between vendors.