tl;dr all your suggestions actually make the attack easier, as the attacker can adapt her/himself to the new rule.
Meanwhile, I would like to explain my view of all kinds of solutions that suggest asking users to propagate blocks in a certain way and asking miners to adopt blocks in a certain way.
First, as Gavin said, such a countermeasure is external to the core protocol and is not obligatory. Hence we better not base our security on the good will of users and miners, unless we point out a really strong incentive for them to follow the proposed change.
Second, even if all miners could somehow always distinguish between the honest block and its malicious competitor block, and all miner would always mine on top of the honest block in case of a competition, there is still a reasonable attack strategy for attacker having more than third of the total hash-power.
Third, I believe we cannot distinguish between honest and withholded blocks, hence the best we can do is to propagate all competitive blocks and let each miner randomly choose on which one to mine. Any other criteria will actually help the attacker and makes thing worst! I have seen three sophisticated proposals on this thread (of ByteCoin, HanSolo, and Peter) that ignores the fact that the attacker can easily adapt her/himself to the new rule.
ByteCoin suggests always preferring the block with more transactionseconds. This way the attacker can simply copy all transactions of the last block in the chain, add some more transactions oh her/his own, and create a block that is about to replace the currently last block in the chain. The honest nodes in the network will actually think that the honest block is the malicious block and vice versa!
HanSolo suggests always preferring the block whose timestamp is closer to the time it was received. This way the attacker can win the race simply by having a better network connectivity than a small honest miner (meaning, having many slave nodes all across the network that propagate the attacker's blocks immediately, without the verification which cause much of the delay).
Peter suggests always preferring the block that takes the least total sum of fees. This way the attacker can simply omit a single transaction out of the last block in the chain, and form a competitive block that is about to replace the currently last block in the chain.
Please remember: block discarding strategies doesnt necessarily involve keeping mined blocks in secret. The underlying principle of the block discarding attack is to increase the attacker's portion of the total number of eventually confirmed blocks, so that it will be higher than the attacker's portion of the total hash-power. This is done by artificially forking the chain: instead of increasing the chain's length by mining on top of the last known block, the attacker sometimes mine on top of earlier block in order to discard one of the honest blocks and replace it by the attacker's block.
Currently the convention is to prefer the block that is firstly received, hence the attacker strategy involves keeping a block secret and releasing it as fast as possible only when a competitive honest block have been found. If this convention is changes, the strategy would change. A more comprehensive analysis of Block Discarding strategies is given in my paper.
In conclusion, the only possible strong countermeasure involves punishing forks, because artificially creating forks is the core principal of the attack, rather than any specific property of the mined blocks.
Happy New Year,
Lear.