For the record:
I'm on the "let there be no fixed maximum block size" side of the debate right now.
I think we should let miners decide on the maximum size of blocks that they'll build on. I'd like to see somebody come up with a model for time-to-transmit-and-receive-and-validate-a-block versus increased-chance-that-block-will-be-an-orphan.
Because that is the tradeoff that will keep miners from producing 1 Terabyte blocks (or, at least, would keep them from producing 1 Terabyte blocks right now-- if we have petabyte thumb-drives and Terabyte/second networks in 10 years maybe 1Terabyte blocks will be just fine).
Right now, miners that use the reference implementation and don't change any settings will produce blocks no larger than 250Kbytes big.
So we're finding out right now how miners collectively react to bumping up against a block size limit. I'd like to let that experiment run for at least a few months before arguing that we do or do not need to eliminate the 1MB hard limit, and start arguing about what the default rules for acceptable block size should be.