I don't know that I'd agree that a 1% change per difficulty adjustment is going to be enough.
All of the baked in constants are of course subject to tuning before final implementation. I used 90% and 1% as examples.
Granted.
Upon further consideration, I think that this might be a great way to move the soft limit, but I still think we need to have a (much higher) hard limit. For the reasons that I stated before, and others, a hard limit removes many of the incentives for big players to engage in anti-competitive activities; since it puts a very real limit to the long term effectiveness of such underhanded methods.
As I mentioned before, we could have it so that the current hard limit is the least of several hard limits. Therefore, we could add in expiring hard limits that get removed after a certain block, allowing us to set it to a new value manually via soft-forks after that point.