The module should return an empty sitemap, or sitemap chunk in case of error

I am not sure the search engines would take a web site without links in the sitemap as a not worth to check web site.

The actual code returns a 404 error even in the case the table used to contain the links is empty; in such case would be better to return an empty sitemap, which is the most correct interpretation of the facts.

Also, the 404 error is interpreted from the users like a permanent error; in other words, they think the error will be returned until somebody will fix the code, or they are said what to do to remove the error message they see.
If the project would return an empty sitemap, we could tell them be patient; the sitemap will get populated with time, which is exactly what happens (taking in consideration that the single modules populate their database tables at cron time).

Also, the 404 error is interpreted from the users like a permanent error.

That is also the assumption taken from the search engines, which will report the 404 message error in the page they offer for the webmasters. In such cases, the webmaster page suggest to verify that all is working, and set correctly; the search engines don't think oh well, let us wait a little more and see what happens, but report the error immediately after they get it.

As I said, in the case of the project modules installed from few time, it's only question to wait the database tables get populated; returning a 404 error causes everybody to think something is bugged, and they need to fix it in someway.

The topic of the report is a little different.

xmlsitemap_output() calls drupal_not_found() when it reveals an error. As the 404 error is interpreted like a permanent error, I suggested that the function would simply return an empty chunk, or an empty sitemap because, most of the times, the error is caused by the xmlsitemap table not populated. This happens because the single modules don't updated their tables immediately; the tables are populated when a node is created/update, when a new taxonomy term is created/updated, when a user is created, or when hook_cron() is executed.
Before that happens, the single tables don't contain data, and the central table cannot contain data either.

EDIT: I had my coffee earlier; that is the reason I can think more brilliantly than my usual. :-)

This is the proposed patch.

EDIT: The proposed patch doesn't work; file_transfer() can be used for files contained in the file directory, and not for files contained in module directories.

This has been changed in CVS.