Gavin's Cost is independent of the trade value of bitcoins, so that wouldn't prevent a halving from improving the attractiveness of fee paying transactions. Regardless, there is much that can yet be done about Gavin's Cost before the next halving anyway, so I don't think that Gavin's Cost is really a problem so much as it is (and will continue to be) the real blocksize limiting factor whether or not we continue to hold a hard limit on blocksizes. With 'thin' blocks (block headers & merkle tree only) the one meg hard limit would easily permit an average transaction rate of 30 - 40 transactions per second right now, doing nothing further. A blocksize limit of 10 megs would get us into the 300+ per second range, which is a workable rate for a major system; and higher rates can be handled better by overlay networks anyway.
Ummm... yeah, I just realised that this
1 - e^(-(1/600) * X)
is not a linear function. This was the source of my misplaced concern. If we reduce a transaction down to the size of it's hash, this cost becomes extremely small.
Thanks for the feedback.