The issues around mixed content took me some experimentation to remember, so I thought I'd add this note to save others time.
When your markup is something like: <div><p>First text.</p><ul><li><p>First bullet</p></li></ul></div>, you'll get XML_ELEMENT_NODEs that are quite regular. The <div> has children <p> and <ul> and the nodeValue for both <p>s yields the text you expect.
But when your markup is more like <p>This is <b>bold</b> and this is <i>italic</i>.</p>, you realize that the nodeValue for XML_ELEMENT_NODEs is not reliable. In this case, you need to look at the <p>'s child nodes. For this example, the <p> has children: #text, <b>, #text, <i>, #text.
In this example, the nodeValue of <b> and <i> is the same as their #text children. But you could have markup like: <p>This <b>is bold and <i>bold italic</i></b>, you see?</p>. In this case, you need to look at the children of <b>, which will be #text, <i>, because the nodeValue of <b> will not be sufficient.
XML_TEXT_NODEs have no children and are always named '#text'. Depending on how whitespace is handled, your tree may have "empty" #text nodes as children of <body> and elsewhere.
Attributes are nodes, but I had forgotten that they are not in the tree expressed by childNodes. Walking the full tree using childNodes will not visit any attribute nodes.