But frankly, if I was dealing with sources in various encodings, I would try really hard to get them all in Unicode before trying to hack something like this.

You may be right, but I don't think it's much of an option for me. This XML system is an add-on to an existing web-app which is 8-bit clean by design. Basically, by the time I'm interested in doing XML I/O the source character set is long gone. Modifying the app to somehow intuit the character set on input is possible, but far from ideal.

Thanks,
-sam