Getting a rough quantitative estimate for the benefit ought to be straightforward in this case. As you implied, there is plenty of (distressing) historical evidence to analyse. Presumably this sort of thing has been done for other safety equipment. I know that when it comes to such things as "g-override", the potential benefits are indeed considered.
Where software is involved though, it's a sod to put numbers on hazards if the software doesn't have a proper pedigree. Basically unless someone uses a very rigorous approach to prove the software works properly, I'd assume it could do whatever it damn well likes, whenever it damn well likes. Including behaving as good as gold for years and then falling over simultaneously in dozens of aircraft when they all do something not exactly like anything the software's ever seen before. OK, so it's only a warning system and ought to only be a benefit or at worst neutral. But in principle, pilots could have modified their own behaviour to increase the chances of needing it to work on the occasion that it doesn't. I can think of one in-service system where that's happened even though it comes without a solid guarantee it will always work in the regime it's been used (although it's designed to and there's lots and lots of experience of it working just fine).
I'm not involved in GPWS, so I'd think that in order for pilots to trust it enough to increase the risk significantly, it would already have established a history of use including rescuing a few close shaves (and so getting a positive balance sheet). A system I am familiar with is in a similar situation, and part of the answer is to be totally rigorous with pilot-applied limitations and procedures. We can do that in a flight-test enviroment, but can we rely on really accurate feedback of how pilots are interacting with a system in squadron service to adjust those limits and procedures as required? It's not that people don't try, but it's not easy.
Sorry for the somewhat disjointed ramblings, but some of use are trying to help and I thought it worth saying so.