Their utility is apparent if you want to just cut numbers down, but I'm not sure they automatically produce the best results if we're hoping to get all the people who could "get it".
My college weed out course experience (20+ years ago) was the first programming class I ever took. It was a C class where a dude read from the book. There was limited to no other resources outside books / internet was limited then. It did it's thing, there were fewer students by the end. I only rediscovered that I actually did like coding decades later.
The varying quality of college courses I think also kinda prove that point. It's awfully easy to say "well it's a weed out course" and just make a crappy course.
But I'm 100% with you on some way of filtering and maybe giving them most of their money back. Granted that last part ... that's going to run into the business folks call and they won't want to do that.
University courses for programming generally get around this issue by having a CS110 type course that functions as a weed out class where people can find out if they have the ability to do the basic problem solving and logical thinking to succeed in that path or not. I imagine it's hard to implement something like that as part of a bootcamp though. The bootcamp really should be pre-screening people using some basic testing or such, but often they are more commercially minded and willing to accept students that will obviously fail because it keeps a steady supply of cash coming in.