the only thing I can thing of is that true/false are (or have the potential to be in this context) expressions, which means one could use compile-time logic to specify inlining. The same wouldn't be true of arbitrary identifiers.
That's why the proposal was for strings, not arbitrary identifiers.
