Incorrect. The block with the best target (and target history sum, more specifically) is the one with the most work. Having a lower hash is just chance.

The protocol is that the first valid chain greatest sum work (and work is defined as the target specified in the bits header) is the correct one. What you are suggesting would harm convergence, especially in an adversarial model— as CJYP notes.  Imagine nodes choose the lowest hash in a race. Say you find an unusually rare block, since you are sure you'll win any equal length race instead of announcing the block you keep it secret until you hear competition in the network.

Even absent adversarial miners, lowest hash is less stable and less safe for 1-confirm transactions since the network is not synchronous some miners can just be late to report and late to switch. Right now a few seconds after you've seen a block and not seen any competition it's fairly likely that the block will not be orphaned, but with lower-hash-wins it would be less likely, moreso when you consider miners wouldn't have anywhere near as much incentive to optimize for block forwarding.

 Miners don't actually give a darn, otherwise they'd do the thing P2Pool has done for years and would setup the ability to relay blocks taking advantage of the transactions sent first. (Which you can get for all blocks running the relay node client, http://sourceforge.net/p/bitcoin/mailman/message/32676543/)