What's interesting is that there seem to be two fairly strongly divergent viewpoints on this matter: some people assume the transaction network will continue to grow to rival paypal or even credit cards, and see the block size limit as an unimportant detail that will be quickly changed when needed. Others see the limit as a fundamental requirement, or even dogma, of the bitcoin project, and view the long term network as mainly an international high-value payment system, or the backing of derivative currencies. Both views seem reasonable, yet mutually exclusive.
It's an issue that will become clearer with time, I think.
By the way, I'm not assuming that Bitcoin will grow to rival PayPal or credit cards. That would be a wild, runaway success that would mark a major turning point in the history of money itself. And the internet is littered with the carcasses of dead attempts to revolutionize payments. It'd be presumptuous to presume future success.
However,
if transaction volumes do grow to reach the block size limits, I would (for now) be advocating for a move to a floating limit based on chain history.
To recap: the primary arguments against are
1) Rising requirements for running a node make Bitcoin more centralized
2) The economics of providing for network security when block inclusion is free and inflation has dwindled
For (1), Satoshi always took the position that Moores law would accomodate us. I wrote the Scalability page on the wiki to flesh out his belief with real numbers. As you can see, even with no improvements to todays technology at all Bitcoin can scale beyond PayPal .... by orders of magnitude. It requires nodes to run on server-class machines rather than laptops, but many already do, so I don't see that as a big deal.
If Bitcoin ever reaches high traffic levels, CPU time, bandwidth, storage capacity ... all very likely to be cheaper than today. I don't think the "only banks and megacorps run full nodes" scenario will ever happen.
For (2) I have proposed building a separate p2p network on which participants take part in automatically negotiated assurance contracts. I think it would work, but it won't be possible to be truly convincing until it's tried out for real. That in turn requires:
a) That there be some real incentive to boost network security, like semi-frequent re-orgs leading to spends being reversed and merchants/exchanges losing money. Inflation will likely provide us enough security for the medium-term future.
b) Somebody actually build the tool and get people using it.
Then you have to wait and see if community participants step up and use it.
In short, I can't see this question being resolved before we actually run up against the limit, which is unfortunate. I wish Satoshi had put a floating limit in place right from the start. But unfortunately there were many issues for him to consider and only limited time to consider each one, a fixed size limit probably didn't seem like a big deal when he wrote it.