It's a strange argument on the face of fit, but nonetheless true. It's easy to go grab the first module you find, and not realize that it's not been actively maintained, might have been superceeded by another module, etc. There's a lot of stuff on the CPAN -- that's good. But there's a lot that lain fallow on the CPAN -- that might not be so good.
Your screwdriver analogy presupposes that all your screwdrivers work equally well. But what if you couldn't easily see the tips before selecting them? You can guess how big a screw one might be used for based on handle size, but what if you needed to pull each one out partway in order to make sure that the tip wasn't bent or stripped?
There's no reason to throw anything away, but there ought to be something that prevents people from having to manually wade through a toolchest trying each tool in order to find what works. Perhaps a voting/rating system or something, not too sure what the solution is to be honest. But I've always felt the CPAN was a double-edged sword, expecially for newer users. Quite a few wheels have been re-invented there. And now I'm out of metaphors. :-)