Brian Beckman and Erik Meijer are well-known figures and often turn up in interesting Microsoft Channel 9 videos, but their latest exploit is written up as a post on the ACM website.
You could even use multiple divs to render a 3D scene. The problem is that modern graphics pipelines tend to work in triangles and a div is a rectangle. The solution is to make use of a quirk of the div that does allow you to show a triangular area of color. Essentially the trick is based on using the border style to create a border so big that there is nothing left in the middle of the div:
Next you shrink two of the triangles, by setting them to zero width borders, and finally set one of the remaining two borders to transparent. The result is a div showing as a single triangle.
The problem is that this is a right-angled triangle with horizontal and vertical edges. A 3D rendering pipeline usually needs general triangles i.e. a triangle between any three points with any angles. The solution to this is simply to break up each of the general triangles into combinations of right triangles. This sounds easy, but it is quite involved - see the original article for the full detail.
The triangle trick was probably invented by Jeff Lau: http://www.uselesspickles.com/triangles/, but to use it to render the complete classic teapot is a work of glorious lunacy. As the authors say:
"The final ingredients of the graphics pipeline are backface culling, Z-ordering, orientation by quaternions, and a kinematics library for spinning the object. Add a little spring-and-dashpot physics, and you can make your teapot bounce, morph, shatter, and unshatter."
It all raises the question of do we actually need HTML5, let alone canvas or WebGL?