[1] That only applies if miners are on average mining break even. With all those Chinese miners who are basically mining for free I presume on average miners are mining with a profit. You are right about the remainder. Worst case scenario the hashrate drops so much that you would need a emergency hardfork to adjust the difficulty.
[2] Not entirely true, the hashrate "ramped up" in the summer, but you can see a decrease from around July 2012 until and somewhat after the halving (which was end november 2012 if I recall correctly).
https://blockchain.info/charts/hash-rate?showDataPoints=false&show_header=true&daysAverageString=1×pan=all&scale=1&address=[2] You shouldn't use logarithmic data when we're looking for a fall on the linear scale. It's even more helpful to use the averaged option as well:
https://blockchain.info/charts/hash-rate?showDataPoints=false×pan=all&show_header=true&daysAverageString=7&scale=0&address=For example if number 1 happened we would see a fall to about 600kGH... Out of curiosity I wrote a script to search for crashes and found a few. It's calculated using the average 14 days, then comparing the new value.
These are all over 25%:
Average Diff New Diff Percentage Decline
9.60447475176 7.14510736828 0.256064745554
536.872239933 367.392898201 0.315679092951
10659.1147545 7641.93145599 0.283061339336
9484.62097297 6860.53271228 0.276667699023
25941.2798106 18975.3491722 0.26852686873
34557.2006617 25621.1386579 0.258587554337
314324099.393 229513533.571 0.269818846169
Most of these were from the very beginning. Really the only one that I'd call a crash is "13/01/2015 18:15:05"
EDIT: Disregard all I said about [2], apparently I have terrible (chart) reading comprehension late at night. I was looking at 2011 instead of 2012. Eduffield (and you) was (were) right and I stand corrected.