I meet the same problem and I skip it when I need a <h2> first for example by wrinting empty tag.
<header level="1"></header> <header level="2>my texte</header>
mentions that the text processing is loosely based on xhtml 2.0. maybe there should be options for the xml field to determine which transformation will take place when the field is submitted to the xml parser. With this we would understand the mechanics better.
I believe Paul Borgermans and his team are working on things like this: