The normal practice is to store as much as possible in binary formats, and store text when required in "invariant locale" format (which just happens to be US English for historical reasons). This becomes even more important for data interchange, and even more so when sharing data via web services and other techniques with global reach.
But for purposes of data entry, display, and reporting you probably really do want locale-aware formatting and value parsing as well as invariant-locale versions of the same things.
It doesn't help much that the terms casting and conversion are frequently treated as interchangable when they are two different things. Oh well.
In B4J we seem to be relying on implicit conversions all over the place, which I personally find disturbing. Right now I have no idea whether these are locale-aware (which could break a lot of programs) or locale-blind using the invariant locale. And I have no idea how to do explicit conversions using the invariant locale, the current locale, or a specified locale.