http://cold-mountain-3633.herokuapp.com/ (tested only in chrome)
1. Does "hover and update only one poster at a time" right. If goodfli.ms had built there website as something that communicated via json instead of HTML, I would have rolled this sort of page into the graphic as well:
http://goodfil.ms/film/89441-cars-2
2. Attempted to make a visual difference between netflix and itunes. I couldn't figure out how they determined whether something was in itunes, so the colors are more of a proof of concept than anything meaningful.
3. Rolled out a trick using fisheye distortion to allow people to explore density spatially via their mouse. To learn how it was done, think about how this works:
http://bost.ocks.org/mike/fisheye/
Overall, this is quick, dirty and probably missing the point of the original graphic.
Self serving plug: if you, fair reader, have data you want to be smeared in d3.js lipstick, I do requests: https://www.odesk.com/users/~~80bea7ba2750c34b
1. Differentiate netflix/itunes by color or marker.
2. Find a better way to show density. Messing with the opacity would help, as would contrasting colors. Nothing fancy like 3D, just a way to convey the density of points.
3. K nearest neighbours is counter intuitive at first. A better layout/ui could have been that when you hover over a point, it shows up to the right as one big poster, without any neighbours.
You have good data that is trapped in that scatter plot.