I don't think there is any need to abandon PO as a format for translators to work in. My main point was that if the text is going to be moved to the database (as per the original suggestion) then an online translation tool is an obvious next step (see below). We would have to write import code to import the existing translations anyhow!
As for XML, I just wanted to point out that a standard already exists, and we should avoid making up our own (if we choose to support XML import/export) unless there are good reasons. I don't have much knowledge (or opinions) about about PO vs. XLIFF, but http://mail.gnome.org/archives/gnome-i18n/2003-October/msg00032.html seems to have some interesting points.
Seems like you really don't know how things are done now. We already store translation strings in the database, in fact original English strings are stored a multitude of times along with translations. And then we already have an import/export facility for PO files. Note that once we started to support PO files (instead of using a web interface), the number of translations dramatically increased, very much passing our expectations. The fact that XLIFF might be better for someone then PO (explained on the link you posted up here), does not mean it is better for us. Someone who knows both need to do a comparision. Show us the needs in Drupal which XLIFF fulfills, and are not possible with PO files.
What is not possible now? What is going to be better?
I think you explained it yourself in your first paragraph. The technical hurdles to learning translation tools and CVS are quite high. While there is no problem with what we have now (other than suggested in the Larry's original summary) there are always non-technical people who want to contribute to open source projects, and translation is an ideal opportunity to allow this. I feel that having the option of using a web-based tool to allow this is good. If I18N is important to Drupal (and it obviously widens the potential market) then the volume of translation will have to increase exponentially, and hence the more accessible we can make the translation the better.
I have seen the number of translations increasing once we started to provide a PO import/export interface, despite the CVS storage. Sure there were translations imported by Gerhard and others for the translators. Going on a completely different route does not solve current problems in itself, in fact it might be harder fo people to adapt. Goba