Scala parser combinators vs ANTLR/Java generated parser?

I am writing an expression parser for an app written mostly in Scala. I have built AST objects in Scala, and now need to write the parser. I have heard of Scala's built-in parser combinators, and also of ANTLR3, and am wondering: which would provide better performance and ease of writing code? So far:

ANTLR pros

1. Well-known
2. Fast
3. External DSL
4. ANTLRWorks (great IDE for parser grammer debugging/testing)

ANTLR cons

1. Java-based (Scala interop may be challenging, any experience?)
2. Requires a large dependency at runtime

Parser combinator pros

1. Part of Scala
2. One less build step
3. No need for a runtime dependency; e.g. already included in Scala's runtime library

Parser combinator cons

1. Internal DSL (may mean slower execution?)
2. No ANTLRWorks (provides nice parser testing and visualization features)

Any thoughts?

EDIT: This expression parser parses algebraic/calculus expressions. It will be used in the app Magnificalc for Android when it is finalized.

Scala's parser combinators aren't very efficient. They weren't designed to be. They're good for doing small tasks with relatively small inputs.

So it really depends on your requirements. There shouldn't be any interop problems with ANTLR. Calling Scala from Java can get hairy, but calling Java from Scala almost always just works.