The object keeps sliding on its background surface as long as at least one of its pixels is visible, unless it encounters a collision that cannot be resolved otherwise. Once that's no longer the case, it pops forward (keeping the same spot under your mouse) until it is.
There are 2 other videos where hopefully you can see this happening. Or, better yet, you can try it running it yourself on Windows or Mac.
Wouldn't drawing with UIKit be less efficient than using an OpenGL backend?
SDL on Windows for instance, falls back to GDL when proper DirectX support isn't available. GDI is what most GUI apps use to draw their elements on Windows. It is akin to UIKit on iOS. But GDI is horribly slow as all the computation is done on the CPU.
In fact, realizing this fact, Microsoft has recently developed Direct2D, which does the computation on the GPU and which regular GUI apps can use to achieve better performance. (BTW, I believe Direct2D is somewhat similar to Cairo 2D.)
 Simple DirectMedia Layer -- used by plenty of games to achieve cross-platform compatibility. Most games in the Humble Indie Bundle use this.
Yep. While UIKit is going to draw using OpenGL anyway with this case you're missing a big optimisation in being able do batch drawing of the tiles and using VBO's etc.
UIKit is fantastic for normal UI's and the 'general case' but almost certainly isnt optimised for doing efficient drawing of a large number of sprites / tiles (trading efficiency for flexibility).
Couple years ago I actually wrote my own API compatible version of UIKit but backed by OpenGL , and then extended support for these type of cases. As it was basically a drop in replacement I was able to confirm UIKit (at least at the time) was not very clever or fast when it came to a large view hierarchy. So my fraemwork was optimising for large batch drawing from a single texture mostly which is what you really want in the OP's case.
I should release it one day as open source.. though is all OpenGL 1 so a little outdated on iOS these days.
This is an admirable effort, but I would not recommend ever using this in practice. I am saying this from experience: I wrote a game engine on top of UIKit, which I spent months trying to fine-tune before finally ditching it and re-writing everything to use Cocos2d.
The performance aspect is just not there. Once you get to a game with a large number of sprites and animations, animations will jitter and glitch. You're really going to be stretching what Core Animation can handle.
What I usually recommend these days is to use Cocos2d for your main game view, but using UIKit for the rest of the interface (menus, score screens, etc.).
Cocos2d (at least when I was using it) engendered some really bad anti-patterns in regards to memory management and architecture. MBTileParser doesn't just offer a way to show graphics onscreen, but it considers the MVC architecture in a way that cocos2d doesn't. (When I started, cocos2d was somewhere between 0.9 and 1.1. The state of the engine has very likely improved since, but that's why I wrote this.) Cocos2d didn't even support ARC back then.