A note on the difficulty:
What it seems to me is that the S4's are on the market and people everywhere are dumping their old miners. I came across a post from 1l1l11ll1l last night (
https://bitcointalksearch.org/topic/m.9110299).. As people dump their old miners for the new ones, or just dump old ones for power issues - there is a significant amount of power leaving the pools.
Granted some of it is being replaced but I'd argue that a lot of it isn't.. Combine that with people who used to mine Bitcoin but have decided to switch to scrypt and I'd say that these two factors are what has been driving the difficulty down.. As more people pick up the old S2's and S1's the difficulty will increase again but it might take a couple of weeks.
I myself just picked up a S2 for $350 on ebay and in doing so I've taken about 243 gh/s offline. And the person that sold it to me took about 1000 gh/s offline.. It's the ripple effect of new miners taking over - it will go back up but right now all the miners are playing musical chairs with their hashing power.
I think this explanation is not plausible at all, because the only reason to take ASIC machines offline is because of difficulty increase putting expected revenues below power cost, but the very thing that causes difficulty increase is putting newer faster more efficient machines onto the network.
Furthermore people who resell old hardware are selling to people who will use that hardware to mine, in no way does that represent "significant amount of power leaving the pools". Even with the delay in shipping during which the old machine is taken offline, there are way too many stars that have to align for these effects to be in coherence such that you actually see a systematic change in difficulty, as there is no reason to believe that these are not randomly-distributed events.
If you do some napkin math on the specs of various ASICs since the very beginning of ASIC mining up to today, you will see that by the time a given generation/fab/process node falls below the threshold efficiency to mine ahead of electricity cost, every single piece of mining hardware in existence at the time that unit was deployed will only come out to a single-digit percentage of the current total hashing power at most.
The only effect that can really drive a declining network hash rate is bitcoin price dropping at a rate that efficiency increases in hardware cannot meet.