TOAST & performance with lots of big columns in a table

I've got an articles table where I want to store texts, of which several translations
exist. Thanks to TOAST I can now store texts of arbitrary length directly in the table,
which is already a big advantage over stuffing them into the file system and trying to
keep the database and file system in sync. What I am wondering is:

From a conceptual point of view, it appears better to keep all translations in one table.
I forget what exactly the argument is; it has something to do with normalization theory.
Anyway I've already got meta information about articles that applies to all translations -
such as author, position within the overall structure, related articles etc.; so if I were
to have a table for every language, then every article row in any language-table
corresponding to a particular article would have to link with the same row in authors,
index, etc., and the structure would get more complicated than it needs to be. However,
with a long article of maybe several 100 K, and translations in 6 languages (this is
theoretical, actually I have only 2 at the moment), the row size would increase
accordingly. Does this pose a problem for TOAST? Is it a better plan to have a separate
table for each language?

- Frank