Drawing back faces in a scene like that would be pretty trivial, if it was done in 3D. A lot of the time it is just better to throw things at the graphics card and let it sort things out, rather than worrying about the fine grained details. I could imagine such optimizations might make sense on low-spec hardware, but that probably isn't the average case.

If you watch closely, the character later on in the video appears to be polygonal by the way it rotates and moves smoothly -- its possible that it was animated that way, but fairly unlikely. If the character is polygonal, it stands to reason that the world is "real" 3D as well. Its also actually quite a bit easier to do isometric in "real" 3D using an isometric projection, rather than to fake it with flat tiles (in "real" 3D, you can use polygonal "tiles" for lack of a better term). When you fake it with flat tiles you have to jump through all sorts of hoops to deal with depth and "tall" objects (or holes) -- in "real" 3D everything just works.

The particle effects on the splashes are square as well, which wouldn't make much sense if they were hand-drawn.

That's just an educated guess though.

It's certainly true 3D, you can tell by the parallax on the tall blocks and, as Ravyne points out, the way the mesh moves.

Though I agree it's most certainly 3D, I didn't see any parallax (apparent changing of relative position). That implies it's an orthographic projection.