[plt-scheme] reduction-semantics with multiple languages

Consider the following definition of two (completely useless)
languages using the reduction-semantics collection:

(module toy mzscheme
  (require (lib "reduction-semantics.ss" "reduction-semantics")
           (lib "gui.ss" "reduction-semantics" ))
  
  (define lang2
    (language (e (x e)
                 (- e)
                number)
             (x (variable-except foo))))
  
 (define lang
   (language (e (x e)
                (- e)
                number)
             (x (variable-except -))))
    
  (define reductions
    (list
     (reduction lang
                (x_ e_)
                (term e_))
     
     (reduction lang
                (- e_)
                (- (term e_)))))
  (traces lang reductions '(- 27)))

The term (- 27) should have only one possible reduction in
LANG. However, if LANG2 is defined as above (and before LANG) the
tracer shows two possible reductions in LANG (27 and -27). The system
obviously uses LANG2's definition of X. Changing the definition of the
non-terminal X in LANG2 to LANG's definitions fixes the problem. I
don't think it's intensional that the definitions of the languages
affect each other. I'm using PLT 208p1. Can anybody tell me what's
going wrong here?

-- 
Martin