Harder parts

Parsed content As soon as an entity reference is included an entity value, it has to be parsed into the document, not just plugged in. (If you set parsed as the default, of course, you may need to protect some entities from parsing.)
Incomplete content XML 1.0 requires balanced content in entity values - no hanging start or end tags. That's a choice that may deserve changing.
Remote content Collecting external data for addition to a document creates new issues. There are performance issues - do you retrieve all the data when the rules file is loaded, or wait until it's needed? Both choices have trade-offs depending on how the processor is running. Similarly, remote content is an opportunity for caching. Security may be an issue, as may packaging.
Complete documents I haven't yet implemented SUBDOC; while you can use this approach to import documents with their own DOCTYPE declaration, I'm not processing that declaration, and any entities used have to be declared in my tool set, not the sub-document. That's a limitation I'd like to erase, but it may not be soon.