The analyzis is flawed, in that it does not take into account all cost. The price of mining a bitcoin tends to approach the price of a bitcoin, so the question is rather is 21's approach equal to or better than large scale mining. If you suppose the chip is free, the design is free, the electricity is free and the management of the device for the mining purpose is free, then it can compete. But none of those things are free.
For the heating appliances then using older lithography makes the mining hardware component of the device nearly free relatively speaking, and the electricity is indeed free because the heat is being consumed.
For the phones, smooth's point has been the chipset makers might be able to fit this onto excess or underutilitized silicon. The electricity is "free" because the users don't count such small portions of their electric bill (and they charge at relatives' and friends' houses too, everything is open house and more shared in the developing world).
The design and mgmt cost can be amortized over this huge scale of billions of unbanked users in the developed world who are eager to have a smartphone.
The competition will be essentially nil, because of the economies-of-scale and cartel level relationships (e.g. Samsung, telcoms in every nation so the device can always send mining shares with no user interaction) required to pull this off in an opaque device that just works on auto-pilot. Once you have eliminated the other miners, you can raise txn fees to any level you want that the market will bear (return of the credit card companies!) and even adjust difficulty to your desired level without a fork by modulating the number of devices you sell into the market
(think it out).Thus the valuation looks fine to me.