There is no loss at all: There are just two parsers for the same Abstract Syntax Tree. After parsing which is the very first step of the compiler, everything is the same.
All the features of Opa lie in the ability to analyze the code and generate JS code from the AST.
But can anyone tell me whether the change from ML-like -> Java-like syntax has meant a loss of composability/expressability/any other traditional strengths of FP languages?