Second, the difficulty is changing by about "last change in % + 5%" to equal the next difficulty that changes about every 10 days or so. This is all according to
http://bitcoinwisdom.com/bitcoin/difficulty. The change is due to the change in the total amount of miners in the pool and the amount of hashing power one brings. That is not the only issue with the change in difficulty, but the fact that new technology with faster hashing power at a better watt usage and cheaper price her GH/s. This will allow each miner to have more shares in bitcoins, but at the same time increase the difficulty.
This isn't true at all. First of all, the "last change + 5%" model is completely off, unless you only consider the last 2 difficulty adjustments (and if you're picking your datapoints so arbitrarily, you can make whatever curve you want to fit work). Secondly, the superexponential nature of this progression is completely unsustainable in every way, shape or form. You estimate the difficulty being a factor 1000 higher in february than it is now. That means that on average for the coming 4 months, every day there is 8 times the current amount of hashing power being brought online. So take all the hashing devices currently mining, multiply that by 8. That is how much needs to be installed every day to reach a factor 1000 in difficulty after 4 months.
Finally, as a minor point. You state that the difficulty changes "about every 10 days or so", which is grossly inaccurate. If the difficulty goes up by a factor of 1.3 (so +30%), it means that the previous period lasted for 14 / 1.3 days. In your model, after ~12 difficulty adjustments, each increase will be at least 100%, which means they happen every 7 days and it only speeds up more in your model.
I agree with the final conclusion that mining is not profitable. But the margin with which this is the case is much smaller than you claim.