Rather than (promising to, and when they don't actually, at least pretending to) use the first-seen block, I propose that a more sophisticated method of choosing which of two block solutions to accept.  Essentially, a miner receiving two solutions at the same height would compute a weighted sum of bitcoin-days-destroyed (transactions received earlier get higher weights) of whatever transactions are in a block and also were in the miner's mempool before the first solution arrived.  Whichever block has more wins.

This strategy avoids allowing miners to use private transactions to mess with the blockchain.  It also makes an empty block far less attractive because it is easily replaced, all the way until the next block locks it in.  Any block-selection heuristic can be gamed, but I believe that using a weighted sum of BTCDD is harder to game than using block propagation timing.

I asked Can Bitcoin Days Destroyed be a better resolution mechanism for competing blocks? on the stackexchange bitcoin site in order to collect objections to and problems with this idea, and have not found any that I haven't addressed.  The best objection is that maybe empty blocks and selfish mining are either good for bitcoin, or else they are so minimally bad that no effort ought to be expended in preventing them.

If anyone here thinks this is a good idea, and no one can offer reasons it's a bad idea, I will probably start working on an implementation.  I'm really slow though, so ping me if it looks like fun to you.

notplato