This is where my conversation with Gavin fell apart. He was not able to acknowledge the concept of a too-high limit. His reasoning was that since the limit was only one-sided (blocks with size above it are prevented) that it couldn't be too high.
Huh what?
...
If you want to continue the conversation, please be very explicit about what problem you think needs solving, and how whatever solution you're proposing solves that problem.
As simply as I can put it:
The proposal is both inaccurate and indefinite, the proposal should not be both.
Now a few words on why...
We do not yet have the mechanism for an accurate limit (one that conforms to need such as those I and many others have proposed), so an indefinitely increasing limit should not be suggested.
It should not be done because it might be good enough for a while, and then suddenly not be good enough and we won't be prepared for that (whether it is too high or too low at that point, we can't say now). It introduces pernicious catastrophic failure potentials.
It may help to think about the US debt limit debates...
The US Congress keeps raising the debt limit. Congress will keep doing it. They wont let it get hit, or if they do, not for long. There is a history of exponential debt growth. So why not propose one that increases exponentially forever? Wouldn't that make sense? It would be practical. There are arguments in favor of it. What it would also do however, is remove the incentive to get it right until it becomes VERY wrong. Its just a limit, congress doesn't have to spend it all.
We have a similar problem with the block chain data size. It is a problem of the commons. The storage and maintenance is done by many, but each individual block doesn't bear the cost. There is a small marginal cost in orphan risk, but this risk fades over time as block rewards increasingly become fees rather than coinbase. There is no economic incentive to run a node, only mining, but the nodes are also necessary for the security and resilience and speed of the network.