How would I do that now? Wikisource pages are not structured data (though Wikimedia Commons image metadata will soon be!), so there is not a clear way to use the Wikisource API to extract just the relevant transcribed text on the page as a field. And on top of that, any text you do extract this way will be full of templates and other code that has no meaning outside of the context of Wikisource. I don't see a way to easily extract just the plain text that is meaningful and relevant (along with other fielded data, like what page or text it belongs to).

Wikisource as a "structured" repository is what we ask from the dawn of time :-)
The problem, as usual, is that if things are left to volunteer developers thing will go slooooowly.
I do think this is fundamental: an ideal Wikisource would ingest and understand many times metadata standards, and would give them back as well.

As for the Wikimedia API, I did this awful script:
Please come and make it better :-D

It just scrapes the data from the HTML (it is localized to it.source, but a quick glance at the HTML source of your own ws could help you, especially if you use microformats) and puts them on a csv.
If you take the HTML you can also get the formatted text.
(I also wonder of a Wikisource which understands Markdown, but that's too far :-)