Think of my suggestion #3 as "proof of history of work."  It is better than proof-of-work because it accumulates history over time of who has done all the work and thus is not subject to sudden takeover.  The longer the network runs, the more resistant it becomes to takeover, since the older miners become trusted, not based on how many coins they own, but how much mining they have done.

In Meni's implementation, weights are given for those who sign special signature blocks.  The weight varies from 0 to their number of coins based on his formulas.  However, his formula seems unnecessary to me.  Why not just base it on the number of coins?  There is almost no difference.  Having a hybrid system means a person would have to accumulate 51% of the coins and 51% of the hashrate.  It's only slightly stronger than the original.  Plus you can't tell when someone is about to combine their coin pile and start an attack.

However, my suggestion assigns a weight based on the number of blocks that miner has successfully helped mine in the past.  That is a different, stronger approach.

There have currently been 235,804 blocks mined in Bitcoin.  If I had been present from the beginning and had helped mine 10% of those blocks, I would have a weight of 23,580.  This is a strong proof of stake (actually, proof of history of work) that always increases over time.  It is a record of all the computer work that I have thrown at bitcoin.

When comparing two different branches, you add up the scores of everyone signing the branch, not just the branch length.  If I sign branch A, and it has 5 blocks, each block has my weight added to it, for a total weight of 23,580 * 5 = 117,900.

An attacker who wants to create branch B would have to create a branch with more weight.  If he is new on the network, he has a weight of 1.  He would have to create a branch that is 117,900 blocks long to overpower the valid branch!

This type of network offers far stronger protection than the 51% hashrate attack of Bitcoin or the 51% hashrate/51% stake combined approach.  The attacker would have to be someone who can override not just the present hashrate of the coin, but the sum of the the entire history of hashrate of the coin.  This is virtually impossible.