This is more of a nice-to-have. The advantages of doing this would be:
  * grouping expectations vs. actual code changes in review systems (both rietveld and gerrit would benefit)
  * would be easier on recipe code isolation logic (wouldn't need to prune out json), and/or allows smaller bundled recipes.
  * tab completion is easier on shells

Concretely, this would mean that instead of:

recipes/
  foo.py
  foo.expected/
    testcase.json
recipe_modules/
  mod/
    example.py
    example.expected/
      testcase.json

it could be

recipes/
  foo.py
recipe_modules/
  mod/
    example.py
test_expectations/
  recipes/
    foo/
      testcase.json
  recipe_modules/
    mod/
      example/
        testcase.json