I just read @rbritt's journal entry about scripting... and I said, check out LuaJIT. If in doubt that's what I'll use for my C++ game. But I really want to roll my own. I've been messing with language design & implementation for 20 years on and off. I've done it before, I can do it again. Might even be practical. But it won't be anything special. So why bother?
Well, coding a simple compiler is nothing compared to coding a substantial game; weeks vs years. Being a game script lang helps keep it simple. And I don't have to worry about a user community; it's just me and my game and a few demos.
Ultimately I hope there'll be languages that are less than ideal, but popular and standardized AND simple enough that any decent programmer can write a compiler that'll run any code written in that language. Forth was kinda like that, and I think we can do better... simple but more mainstream. Unlikely, but maybe my language will be a big hit. Maybe I'll find some other language I like, write a better compiler for it, and suggest a few tweaks. Maybe other people do all that before I get around to it, and I catch a free ride. Doesn't really matter. Any of that would be better than the most likely outcome, that all languages will suck forever and ever, so get used to it. But that's no reason to give up. If this problem is ever going to be solved, who's gonna solve it? Gamedevs!