The "sunken" production cost of "old" bitcoins is quite irrelevant in this aspect, and most of the mining will always happen where the electricity is cheapest - but for Bitcoins to be a useful "unit of account", the production cost should be fairly constant over time - and it is not.

Actually, I'd argue that the author has put the horse behind the cart in his article, electricity consumption is a function of the bitcoin value, not the other way around. The higher the value of the bitcoin, the more energy will be used on mining, and the higher difficulty level. The lower the value of the bitcoin, the less energy will be used on mining.

I'd also argue that it's a quite bad design choice both to retarget the difficulty after 2016 blocks as well as to halve the reward every 210000 blocks, the algorithms should have been more smooth - now both the difficulty retarget and reward halving can be like major events affecting the market price a lot. Ideally, things should have been done more frequent and in smaller steps, and only external factors should be affecting the market price.