[banachtarski]

In graphics, everything is done at subpixel resolution. We also have to handle text and the aliasing that occurs not just from screen resolution downsampling but projective aliasing as well for text that is not camera facing or text that is wrapped on a surface. Keeping track of text on a screen is, to put it simply, not hard. There's a lot more state to a level's geometry and textures than just position. Momentum, for one thing, but tint, wind, pathing, any other behaviors, key frames, etc. Some of the meshes get skinned or dynamically deformed if its cloth or hair. Some meshes get generated on the fly. Still others move according to a script or analytic equation like particle quads. Not to mention audio that may be keyed per item and the physics tick.

The loading is sometimes more than just pulling stuff from RAM. Decompression, shader compilation, and more. Maybe textures are created dynamically or the level is randomly generated (terrain, creatures, and all). Usually, loading is full threaded and all reading happens asynchronously. Depending on whether you're on a console or not, different optimizations happen since the disk drives have different buffered reading characteristics.

I've read plenty of stuff on web browser tech and architecture, particularly because I've been following the Servo project with some interest. I'm not saying it isn't a lot of work (it's a lot of work, like anything worth doing), but I do think it's easier than you're making it out to be.