I thought I'd take another run at fixing this test. The test as it
stands now fails because it checks that some particular preference
changes between Locale changes.

As it happens, it also assumes you have multiple Locales ('en' and
'ja', in particular). Trunk images don't, so the entire test is bogus.

It seems like the thing to do is to have a specific test Locale, and
rewrite the test to
* store the original Locale
* switch to the test Locale
* check in any way possible that the Locale changed (say by adding
LocaleTest class >> #localeChanged with a flag Boolean, for instance)

Sound good?

frank