I've thought about that for some time until it started to feel like a wrong question.
I usually wanted to code in one language to save learning many languages to an expert level. Also it felt good to have a uniform code on various levels of the system - something which today in JavaScript world is called "isomorphism". Then I realized I'm really doing different things in different layers, and in a big language I'll perhaps use different parts of the language - effectively several languages. For a small language I'll probably be creating necessary constructs first - again, different for needs of different layers - and then would use those.
Now I think it's better to have the ability to adjust language after the task, and here Lisps look rather good. Even more, with security in mind I want to use not only "language suited for the task", but "language minimally suited for the task", so it's hard to use it incorrectly and easy to reason about abilities of the particular software layer. Here, again, Lisp-like languages are a good platform. Some others languages suitable for DSL creation (Haskell? Tcl?) can also be good.