I\'m doing some i18n on a web-based app using Django, which uses gettext as its i18n foundation. It seems like an obvious idea that translations should be stored in the databas
It's a very common way to do translations that has been around for a long time allowing any issues to be ironed out over the years. I imagine writing something like gettext it would be all too easy to make incorrect generalisations about how languages work. Why should the Django development team spending time researching that and developing it when it's already been done in a tried and tested system? Furthermore professional translators probably know what to do with PO files where as a home-brew translation database may prevent them from working in ways they're used to.
Why would you prefer translations in a database? I guess you might prefer it as you could make a translation interface to the database. If that's the case have a look at Pootle it's a powerful web-based translation interface that works directly with PO files and can even integrate with common version control systems. Add some post-commit hooks and you can have such a system with little work and without the overhead of a translations database.
Hope that helps.