The <desc> element, which can come after the title (caption) of a table or figure, should provide a fairly detailed description that could go in the summary attribute of a table or, for a figure, perhaps a linked-to file with a description readable by a screen reader for the visually impaired. Although the DITA spec doesn't say so, a good use of the <desc> content would be to provide additional content for the visually impaired, while the introductory description of the figure or table should be sufficient for sighted readers.
In practice, the DITA toolkit just places the text of the <desc> element after the table or figure caption, visible to all. In the frames version of the manuals, there is no line break; in the one-page HTML and PDF, there are line breaks surrounding the description. I think that if the text in the <desc> element is useful to all readers, it should be placed in the introductory text instead of the <desc> element.
The <alt> element is meant to translate into an alt attribute in the HTML img tag, a brief placeholder for an image that appears only if the image is missing for some reason. This should be a short phrase, but often in the Derby docs it is a very long description more appropriate to a <desc> element. Moreover, although the alt text is included in the frames version of the documentation, it does not appear in the one-page HTML or the PDF.
There is no good solution to this problem at present, because of problems with the toolkit. At this point it probably makes sense to leave all figure text that should be available only to the visually impaired in <alt> elements, rather than make them visible to everyone. For the few tables that have <desc> elements, the text really belongs in introductory text (these are at the end of the Reference Manual).
See http://db.apache.org/derby/docs/dev/devguide/cdevdvlp27610.html and http://db.apache.org/derby/docs/dev/tuning/ctundepth32379.html for examples of what needs fixing.