I've never heard of XLIFF but from your description it sounds like a lot of "corporate" software projects - big, ambitious, doesn't do anything as advertised, yet somehow it is supposedly used in "Real Enterprise Systems." It's very common to come across these sorts of projects in the Java and MS ecosystems.
<Reads spec>
I was right. It's underspecified for real use-cases - no coherent model for implementation emerges naturally from the spec, other than "grab the fields you plan to use and ignore the rest" - and it has a ridiculous kitchen sink of optional fields that, in some ideal fantasy land, XLIFF producers and consumers would all support. Then they mention "oh, it's extensible just in case." Which means everyone is going to go and break the spec to get shit done.
The only way such a bloated spec could ever get traction is if someone like Adobe put together a massive app for doing translation work, one that substantially improved the process, and made XLIFF the primary export format. Apple might have picked it up because someone in the company's engineering staff pitched a fit about standards, but I can't imagine they're making much use of it themselves.
If you want to see working examples of game localization, try looking at what open-source games are doing. For example, Wesnoth has some prolific language support now:
http://svn.gna.org/viewcvs/wesnoth/trunk/po/wesnoth/ I had no idea what "po" is - took me ages to find that directory - but it appears to be the compiled result of this tool:
http://www.poedit.net/ Very promising stuff.