I think because the pros don't use it and most wannabes spend most of their time studying what their programming heroes do.
That is true. Collada is very good for Playstation but should it mater to everyone else as much as it maters for Sony and their parteners? Creating a game engine is full of pitfalls so it's good to learn with those that made something instead of just following related game sites.
It's quite ambitious to wish to create an unique format to rule them all... COLLADA isn't adapted for levels.
I never said that and i don't know Collada in deep imideatly discard it. But reason with me. You are not going to set path grids by hand. You need to browse your scene and edit the path nodes. Even if you use a tool to automatically place the grid, tools aren't perfect so everything needs to be visualized and checked. So if you have to make your scene editor anyway why bothering with Collada? You could have an option to import Collada objects for static and animated object but read ahead.
Path grids are just an example. There are toons of other game objects like activators, physics objects, traps, etc that will require a game scene editor.
Collada has its advantages to export animated and static scenes using basic skeletal animation techniques. But if you try to do something more advanced, like "Die By The Sword" did, again, i will have to ignore Collada because it's not useful. So we are left with Collada for static scenes, lights and materials. No wonder that so many people use their own scene and data formats, with versions of xml/text for exporters and binary versions for loading speed.
This is an area where Java can take advantage from the competition. And using X3D instead, because you can create your own personal extensions, and use whatever encoding you want (xml, vrml syntax, text or whatever) may be a better choice for your scene editor.