The short answer is that we do not implement RFC 3023 currently. The RSS platform uses MSXML (in XML conforming mode) to fetch and parse the data, so the behavior is inherited from MSXML. Since MSXML is used by most products that we ship, it means the platform is consistent. And nearly every other stack in the industry ignores RFC 3023 as well, so it's not a widely accepted interop point at the moment.
I find that interesting because the short answer from 3023:
If an XML entity is in a file, the Byte-Order Mark and encoding declaration are used (if present) to determine the character encoding
Is something I implemented in BottomFeeder a long time ago. At the time, I found that paying attention to the declared encoding declaration helped a lot. I guess the MSXML parser isn't built that way, and I'd also guess that the decision was based on the sources they ran into during construction of the parser. I was looking at RSS feeds, and the declaration seemed like an easy thing to do.