EDIT: The more I think about this idea of the bias the more scary it becomes.  It not only just restates the difficulty time warp problem as a difficulty+timestamp time warp problem, but actually gives your average miner motivation to timejack attack his peers without even caring to attempt double spend, just to bias their notion of the current time and increase the odds that a "far out" block will accept.  Difficulty adjustment becomes unreliable and volatile. A whole new family of attack vectors get created.  It is all around scary stuff.

I'm now much preferring the idea of simply returning to a more classic adjustment, allowing difficulty to potentially go "too far" and including an upward pressure on targettime to bring the chain back into a mineable state should the difficulty exceed the possible traversal time of any possible map generation.  Such an approach seems both very safe and very simple now, by comparison!