Interesting thoughts William, however much of what you describe already exists: 

This happens with users deciding to pay (essentially) a fee smaller or larger than the 1999th tx in the mempool. If user is willing to pay more than the 1999th highest fee tx in mempool (or whatever tx byte 1,000,000 in a mempool ranked by tx fee) then they will get to be on the next block. This is a simplification but you get the idea.

Further, Greg Maxwell's one reply covers alot of the biggest pitfalls well. Especially that much of this already happens and that behavior in response to such a change could be hairy to predict. So I'm not sure that---what is basically--- a multi-unit Vickrey auction is the best way to go... but it is an interesting idea worth further examination.  

Second, minimizing invididual user tx fees and maximizing total tx fees are essentially incompatible within the current development structure. Even if the block limit is increased to 3mb the same problem will eventually occur. While I agree it would be ideal to maximize social benefit (i.e. maximize both consumer and producer surplus) through more simple protocol changes---at least in the medium term, that isn't on the table so it is better to move on. 

  This is not an easy optimization problem to solve though. The difficulty is trying to model a market like bitcoin; the only thing that comes close is electricity markets. With 2,000 tx in the pool, users aren't willing to pay shit to send a tx; there is always a miner to process it. So at 2,000tx per 10 min (3.333 txps, 1mb/10min,etc) and below users price elasticity is flat. Once the pool has >2,000 tx in it, especially for any extended amount of time, users price elasticity is about as elastic as a brick and goes near vertical. This creates a situation where miners are always better off when there is a significant backlog (this can be seen in miners revenue from tx fees anytime there is an ongoing backlog). 

Simply put, it would take some very large blocks to have total fees/block exceed total fees/block for constrained size blocks given the near vertical price elasticity users face when there is a backlog.

Users willingness to pay is what they pay---no surplus. Miner elasticity is more or less flat but we can call their willingness to accept whatever the lowest fee tx is in a block. Say it is 180 sat/b (N_L), and the highest tx fee in a block (N_H)is 320 sat/b. If you subtract each tx in the block (N) greater than N_L 

and sum the result you get surplus to miners:  Σ(N_1,2...- N_L) 

So, yes, the multi-unit Vickrey simply shifts the surplus to users. However, a case could be made that since---as mentioned in an earlier reply, the optimal strategy for miners is to accept zero fees (given their flat elasticity), and therefore all fees are surplus benefit to miners---shifting this surplus over to consumers could create some good effects. Primarily pushing users' price elasticity away from near vertical inelasticity as it would take some of the upward pressure off rapidly increasing fees in a backlog scenario.        

It would be interesting to see a simulation of how this would play out, but without that this is too risky to implement.

Regards,

Ryan J. Martin (tunafizz / ohituna)