Currently I generally just don't test this case.  Example: what happens if you run insertOrderedList on <span contenteditable>foo[]</span>?

IE9: <span contentEditable=""><p>foo</p></span>
Firefox 9.0a2: Throws NS_ERROR_FAILURE
Chrome 16 dev: <span contenteditable=""><ol><li>foo<br></li></ol></span>
Opera 11.50: Throws NOT_FOUND_ERR

How about <div contenteditable><span contenteditable>foo[]</span></div>?

IE9: <div contentEditable=""><span contentEditable=""><p>foo</p></span></div>
Firefox 9.0a2: <div contenteditable=""><ol><li><span contenteditable="">foo</span></li></ol></div>
Chrome 16 dev: <div contenteditable=""><ol><li>foo<br></li></ol></div>
Opera 11.50: <div contenteditable=""><ol><li><span contenteditable="">foo</span></li></ol></div>

<span contenteditable><ul><li>foo[]</ul></span>?

IE9: <span contentEditable=""><ol><li>foo</li></ol></span>
Firefox 9.0a2: NS_ERROR_FAILURE
Chrome 16 dev: <span contenteditable=""><ol><li>foo<br></li></ol></span>
Opera 11.50: <span contenteditable=""><ol><li>foo</ol></span>

I see three sensible options:

1) Pretend the editing host isn't inline, and give it block children.

2) Silently refuse to add new block elements if there's no ancestor in the same editing host that they're allowed children of, but otherwise execute the commands normally.  E.g., insertOrderedList should refuse to insert lists but should change existing <ul>s.

3) Bail out of all block-level commands unconditionally if the editing host is inline.

I'm leaning toward (2).  Opinions?