Aren't diffusion values an implicit representation of pre-calculated closest path pathfinding? Calculating diffusion incrementally is a way to update the graph over time with a known incremental cost. But it all is just a variant of Dijksta in the end, isn't it? :-)
It's highly parallel, and suggests also how multiple agents might collaborate to seek a goal (such as blocking all the possible exits that a pac man might take). See section 3.2 of the paper (http://www.cs.colorado.edu/~ralex/papers/PDF/OOPSLA06antiobj...) for an example of this. It's also an approach they've taught with some success to very young students. Seems extremely significant and need-to-know for anyone implementing game ai, even if they've already implemented other pathfinding algorithms. I'm certainly happy to see this! It's awesome and eye opening.
http://www.cs.colorado.edu/~ralex/papers/PDF/OOPSLA06antiobj...