It's an interesting option, but I think it runs into some problems by aiming for more complexity. From some quick anydice-ing:
- 7d6 works best: ~75% of the time you get to roll 14, then run out of dice
- a mix of 7d6 and 6d6 gives you slightly more range, but your chances get worse enough that it's probably not worth it
- anything else means you die in a few turns or waste dice (7d6 gives you a 98% chance of success per roll, 6d6=~90%), 5d6=~70%)
You could make the target less harsh (e.g. >=10) to extend the game a bit.
The real problems for me are 1) it has more and clearer responsibility for failure than jenga (where pulls are more drawn-out, skill-based, and obscure). Basically, it feels more arbitrary, but somehow also more my fault. 2) It adds a gamble, but everything is always at stake, so there isn't really any choice. That, and choosing which block to pull affects every future pull, but choosing how many dice to roll only affects how long the game can go on for.
I mean, these both have something in common (any failure = instant death), so what if there were multiple things at stake that you distribute dice between? That'd give more weight to your choices before and after you lose the things, since you could have pointed oracle questions about them even when they're gone. Alternatively, multiple layers of failure (e.g. someone posted an idea in the idea thread about a journalist on the case of a serial killer, who kills again every time the tower falls).