I don't think it's worth doing if it can't be done on a P2P basis. I don't think anyone will use it if it has any centralized authority, and besides it creates a weakness that can be shut down or dismantled.
Personally I think it would be enough to just prevent the price from rising too quickly. Some way to preempt speculative bubbles by decreasing mining difficulty at the right time, then gradually increase the difficulty back to where it was, such that the net effect is that supply increases in sync with its increase in market capitalization, aka demand. But you're right, that doesn't solve the problem of decreases in demand. I don't think people would be willing to bear any demurrage or fees, but perhaps mining difficulty could be increased to very high or infinity in the event of inflation? This would temporarily cap the supply, which would raise the value, but it would be a lot slower than the other direction. Still, if it could be done I think it would work pretty well.
The problem is, how does one measure a currency's value (or change in value) without comparing it to anything else? Are there any measurable, intrinsic monetary indicators that correlate with a currency's value?
I agree that the value-stabilization mechanism needs to be a part of the protocol and be entirely distributed in nature and not rely on any individual or group to function as a unilateral decision maker. Rather it would be a bit as you describe a counter-cyclical feedback mechanism that expands or contracts money supply as necessary.
In Freicoin we have implemented Demurrage of 5% per anum, with the equivalent amount mined by miners to maintain a stable monetary base after the initial distribution is completed. If the miners rewards were raised above or below the amount removed by demurrage it would effectively increase or decrease the monetary base. We believe this demurrage fee to be beneficial entirely for its effects on velocity and interest rates and we don't use it too adjust monetary base, though its the logical mechanism to use when the base needs to shrink.
I do not believe that hash-rate would be an adequate mechanism, while their is strong indication that hashing establishes a floor to BTC valuation it's dose not reflect well the upward speculative swings. I think the only way to get a handle on the problem is to run a distributed prediction market for the future value of coins inside the block-chain itself. If the market predicts inflation less coins are made, if it predicts deflation then more are made. All users would be able to place bets on either side putting their coins on the line and profiting if they are correct, thus by putting 'skin in the game' we can expect the predictions to be accurate.
Unfortunately I have not been able to come up with a reward structure that works, because all bets and rewards need to be in the the coins of the chain and need to be enforced by the block-chain such that the rewards are correct without any outside verification of the purchasing power. In essence I need a transaction or financial instrument that can be bid upon and then 'won' by a party such that if their inflation/deflation prediction is correct they profit automatically and the person taking the contrary view loses automatically without any central authority deciding arbitrarily.