There's no guarantee, but Satoshi's paper addresses the dynamics of this - rational miners shouldn't want to undermine the validity of their own wealth.
Similar to how rational coal power plant operators and chemical plant owners wouldn't want to undermine the cleanliness of the air they breath and water they drink.
If a coal power plant operator controlled 20% of the coal plants in the world, they might be concerned, as there would be a measurable effect on their health from them choosing a more polluting way to burn coal. This is comparable to the impact that the decisions of mining pool operators have.
Now regardless of the wisdom of trusting 0-conf transactions, the fact remains that no one is forced to accept them, and removing the option by a deliberate change in standard client rules is leaving people who would otherwise choose to risk a double spend on a 0-conf transaction without the option to do so. How can that be rationalized? "We know what's best for merchants" isn't a compelling argument.
By all means, warn merchants that 0-conf transactions can be double spent, and explain the methods an attacker could use, but still let them choose what level of risk they want to bear.
I don't see any need to change Bitcoin here. If you don't think 0-conf txns are reliable, OK, wait for a block. Or tell people who will listen to wait for a block. Those who want to see how reliable they can be made using technical fixes like mempool sync, doublespend alerts, risk analysis and so on can then go ahead and do so. The market will end up deciding who is right. If merchants keep getting double spent they'll go to waiting for a block. If they don't, then we all win with a more useful currency.
+1