The quality of saturation is the critical element one should be worried about.
Saturation from dust, spam, worthless micro-txs, faucets, dice, "stress tests", scripts that are purposefully wasting space in the blockchain for minimal fees, etc = nobody should even bother.
Saturation from legit txs should get a bump to accommodate for new capacity.
That is true. However, to kill that "trash", it would be much better to just raise the minimum fee. It would be a "centralized decision", true, but so is imposing an arbitrary size limit.
If you enlarge the block size prior to being genuinely saturated, you are just opening the bloat-attack-vector, wide open.
When the 1 MB limit was introduced in 2010, it was 100x the average block size at the time. Until the "stress tests" of June 2015, there has been no "bloat-attack-vector": no one tried to create 1 MB blocks, not even in 2010 when they would have been more damaging than 8 MB blocks would be today.
Quite probably, the stress tests would not have happened if the limit had been raised to 8 MB two years ago. The stated goal of the first test was to create a 200 MB backlog of unprocessed transactions. That was easy, because the clearance between the normal traffic (0.5 MB/block) and the effective capacity of the network (0.8 MB/block) was very small, so the "tester" only needed to issue more than 0.3 MB of transactions every 10 minutes to start building the backlog. If the limit was 8 MB, the effective capacity may have been at least 6 MB/block, so the tester would have had to issue 5.5 MB every 10 minutes -- more than 15 times.
For the same reason, the 1 MB limit makes a real spam attack (with dynamic fees) much more likely, because it makes it much cheaper and easier to carry out.
If someone says "but we can send these users to thin clients, web wallets etc", just contemplate that if this happens now, and having your own bitcoin-qt is unsustainable for a lot of people, then what will happen in 5 years or 10 years? such txs.
Centralization of mining is one of the few real flaws of the protocol, that will require another genius idea to solve. The "full but non-mining" nodes are a hack that attempts to get around to that problem. However, they violate the spirit of bitcoin, they will not protect bitcoin from a miner cartel, and they will disappear anyway, because they are not rewarded (and they are not rewarded because they did not exist in Satoshi's design).