The one concession I have been willing to make to the fine chunking characteristic of many DITA implementations is that it was a boon to translation. Apparently not so, according to a recent blog post on Content Rules.
The problem is that fine chunking tends to obscure context, making the content impossible to translate reliably. And the real kicker in this problem is that even if the translator is given the means to see the content in the current context or contexts, the source may be reused in new contexts later without the translator being involved again or ever seeing the content in its new context. (This is where the savings are realized, after all.)