Um... This may be a stupid question, but why not use Java and just code the game-specific extensions yourself?
It's not a stupid question. It's the question I had after reading that. If he's going for public modability (thinking Java is too complex for the public to learn, which it really isn't), but he still wants all of those features, he might as well use python or lua hooks for in-game methods and information. Fairly simple to create interfaces between the two.
From what I've seen it's a little pointless to design a new scripting language for a game when you aren't working with anyone else, because any speed increase you can get from it is usually negligible compared to the gains you have in development speed by using something simpler (i.e, you can spend more time optimizing the rest of the code for larger performance gains than you'd get over the same time period struggling with developing tighter generic methods for use in scripts)
I'd probably guess speed, honestly, but I'm not really sure. I've never seen a Java program of the size or complexity even approaching Inc that ran worth a shit, to be blunt, and Inc's already had some fairly serious resource issues (re: saving, ferex) in the past.
Java is perfectly fine now speed-wise, especially for a turn-based game. Incursion has a lot of stuff to track per-object, but Java is designed for efficient OOP to begin with, and beyond some small overhead from the VM, it can actually be better out of the box at resource management (the garbage collection he mentions is part of this) and memory footprint.
My guess is that he doesn't want to learn a new language to develop his game, which is perfectly understandable, but if he's developing a type of Java on his own just to avoid that, it becomes a bit questionable in sanity. Hell, he might be doing it for fun, I don't know. It's his game. He's probably already considered it to the extent he wants to/needs to, so asking him won't do much good.