As the syntax will not be lisp-like, it will not nearly as easy to write a translator as it would be otherwise, but lisps such as arc are still very concise languages, and macros give you a lot of power.
Which would you rather do? It will probably be challenging either way.
1) powerful macros
2) functional and imperative programing
3) strong tree handling functions
Disadvantages:
1) arc is somewhat slow.
2) arc has few libraries
There might be more, but lisp is just a programming language like everything else. It might be somewhat better, but odds are it won't be a "silver bullet".
I think that Lisp can support good quality syntax parsers. An example of this, although it's not a full fledged language, is CatDancer's JSON parser built on top of a general-purpose parser library: http://hacks.catdancer.ws/parser-combinator-approach-to-json... (it's a long article, the relevant parts are the ones about the parser-combinator approach).
Also, it seems as though your in-house language isn't completely specified yet. One area where Lisp excels is implementing unspecified things, and changing them with great flexibility on-the-fly.
Shader mentioned that Arc is slow, and this does seem to be the case. If you really want to use a Lisp, and Arc is too slow, it's probably worth looking at a few other Lisps before turning to Option A -- there are very fast implementations of both Common Lisp and Scheme.
