There is most likely no need to do incremental changes. There are already pretty good looking suggestions on how we can simply let miners decide the block size. To do that we would need some fairly strict rules for how long validating blocks by regular nodes can take until they are rejected. This would create a cap for the block size that is actually relative to the processing power of most full nodes. That is something we want, I think.
The other interesting suggestion was linking mining difficulty and the block size. Those are the two decent suggestions so far. Finding a long term way to fix this is much better than simply making a one time increase. If there is a one time increase, it should be a massive increase together with a smaller soft limit, so that the soft limit can be raised more easily in the future if necessary.
While I can see your point, I still think that there should be a hard limit, however high that might be. In this way, there is always an absolute for which large miners 'cheating' by trying to force out smaller players (via padding of the transaction queue or similar) will run into that limit. Hopefully this absolute limit will discourage those unscrupulous players from turning to the dark side, simply based on the idea that there are going to be a percentage of players that they could not force out of the mining business, no matter how much larger they could grow on a relative basis.
However, upon further thought, I think that said hard limit should be pretty freaking high, in order to allow the soft limit to be our control method for many more years without further need of hard forks. And if we are going to do this hard fork against the desires of some miners (which is probably a certainty) it's better that we get to it sooner rather than later. If we start bumping up against the hard limit, some miners might just manage to profit from that artificial scarcity in ways that would turn a good portion of the miners aginst the idea of a hard fork, whereas they might otherwise not oppose one before we get there.
How high would such a hard limit be? Can we estimate how many transactions per second that, say, a one Gb hard limit could process? As already pointed out by many, we don't need to accommodate all of the transactions that occur in the world; but we need to be able to do much better than 7 per second. What is a good target? VISA's transaction volume? Paypal's? Or VISA + Mastercard? Once we decide, then we need to set it and let it go, and be willing to let the market in transaction fees and off-network transfers simply run. We are going to get way too big to do this kind of thing twice.