The two you mention (Haskell and Prolog) are great examples of languages where people can actually have negative impact for a while until they really get up to speed with the language. Pretty much any functional or declarative language fits that bill, and definitely any Lisp. Maybe not 6 months, but definitely at least a couple of months to start contributing a lot of additional value (in addition to getting set up with the team's environment and workflow).

Amazingly, even if someone does learn Java, Python, and C, he or she is not necessarily able to generalize that knowledge to other languages, or even to larger-scale projects in those languages. For example, I wouldn't want to take someone with a lot of Python experience and a semester's worth of C, and put them on a large-scale firmware project or Linux-kernel-hacking project. There are a lot of ways somebody who thinks they know C could mess up a project like that.

In general I agree with you: a truly smart engineer can pick things up as they go, especially on a longer time frame. However, this isn't always the case. Though my perception is admittedly warped a bit working in startups, where you need people to hit the ground running immediately.