I like that idea. But of course developing and marketing that capability would take time, and so would be another kind of risk (albeit one that could pay dividends for all future games).
I'm not familiar with Jeff's code, but I do work in game development, and something like this would likely be a huge undertaking, and would affect pretty much every aspect of the game.
Optimisation #0 for loading times is compress a gigantic pile of stuff into a giant binary blob, and index it at runtime. Undoing that to support custom content would be a huge time investment on any of the games I've worked on.