Because mining is dependent on the rate at which information can propagate across the network. If a mining pool is able to hear about the next block 1s sooner than another pool, it has a significant advantage when blocks only take 12s to mine. If blocks take 10min to mine, that 1s is less of an advantage.
If miners get paid for valid blocks that don't end up included in the chain, that also makes the 1s less of an advantage. Hence GHOST. Here's the original paper [1] and a later one [2].
Sure, maybe including uncle blocks can help secure the chain. It's not without cost, however: now you're forever persisting valueless blobs of data as part of your blockchain.
It's just an extra block header every few blocks, which is a minuscule amount of data compared to the transaction records. And it's not actually valueless since the uncles are part of the block selection algorithm and therefore contribute to security.
