I think the hardest part about running such "experiments" is ever knowing whether they were success or failures. Even here, we (and I doubt CDPR) don't know enough to really judge whether the tokens things helped or hurt. Or if it helped, whether it helped as much as a more traditional system. Except for small experiments that can be run many times (or in many similar departments in parallel), circumstances just change too much to ever compare apples to apples. (Here, devs may have loved this game's concept so much that they worked hard in spite of the tokens. Or didn't love the game concept but loved the tokens.)

Frankly it's why it doesn't make too much sense to veer very far from practices proven to work. But what's shown to work in a HBR studies may not apply in a different cultural landscape than the US.

All that being said, I'm sure there are many possible practices that are better than traditional approaches. It's just risky trying to find them.