Using prevIndex in Delaunay.find is efficient when the new coordinates we're "find"ing are close to the previous one (like, displacement of a few pixels). For a couple of random coordinates, I don't think it brings any improvement.

Ah yeah, good point!

And while you’re here… do you have any thoughts on using Delaunay vs. the bisector method?

bisector is nice… but I think the fastest method would be quadtree.find?