Eventually there won't be enough profit in it and mining will be done by those who have high transaction volume.
Bitcoin mining will evolve to a market where it's not just about transaction volume. Some miners will ask for a higher fee and some for a lower one. The ones who ask for lower fees get higher volumes but have a smaller profit margin and vice versa.
My point is that some miners will survive even without high transaction volume if there are enough users who want to pay a higher fee to make sure their transactions will be included in the next block.
To znort987 and others:
Personally I don't believe Bitcoin mining will ever become excessively centralized, for many reasons. One of the reasons is that some of us have fixed electricity costs, it's almost impossible for any large scale operation to make mining with fixed electricity cost non-competitive. It is not realistic for a large mining operation to be able to mine that cheaply even if they manage to somehow acquire the hardware itself for cheaper than the regular joe.
In Finland there is often a fixed electricity cost (included in rent) for student apartments, but those are usable for small mining operations only. You have to also take into account that many small time miners do not have any extra costs for the space, cooling systems etc. For a large mining operation that is a totally different story.
Then there is the fact that some people actually heat their cabins or houses with Bitcoin mining hardware these days. The added benefit of the heat makes it very difficult for the mining to become overall non-profitable. I believe a large percentage of our current mining power is people who just have their desktop PC's video card mining bitcoins or something like that. Then another large percentage of small to medium size mining operations. Large operations are a small minority of the total hashing power.
This will change in the future though, large operations will have a bigger share and the "single miners" a smaller share but I find it difficult to believe that small and medium sized mining operations can be ever driven away entirely, especially if they have the edge of fixed electricity cost or no extra cost for space and cooling, or if they are using the heat for something useful. I'm currently mining with fixed electricity cost and I have no additional costs. Mining profitability would basically have to be bad enough that it doesn't beat my hardware's rate of decline in resell value or the risk of hardware failures, for me to stop.
Going FPGA and then ASIC doesn't change anything either, I'm already considering investing in FPGA's. They are actually much more convenient in many ways compared to a traditional rig. Bottom line is that Bitcoin mining is not that different from any other modern business. Large operations will have a bigger role in the future but small and medium sized operations won't go away either.
I also wouldn't make large scale mining the devil. It's important to understand that running an operation like that involves much bigger risks as well. They have the biggest incentive of all to keep it running smoothly and profitably. That is why I'm not really worried if the Bitcoin mining network becomes a bit more centralized than it is now. If it does turn out to be a problem, I bet there will be solutions as well.