because actually i can do :

<dl>
<dt> 1st terme </dt>
<dd> 1st terme description </dd>
<dd> 2nd terme description </dd>
<dt> 2nd terme </dt>
<dt> 3rd terme</dt>
<dd> 3rd terme description</dd>
</dl>

with no way for AT to know that 2nd terme description is actually the 2nd terme description and not the second description of the first terme

Actually, no, you can't. What you're doing is not conforming to either HTML 4 or HTML 5.

http://hsivonen.iki.fi/validator/html5/?doc=http%3A%2F%2Fwww.fairytells.net%2Fdltest-html5.html
http://validator.w3.org/check?verbose=1&uri=http%3A%2F%2Fwww.fairytells.net%2Fdltest.html

both validate the page

the html 5 spec say :
bt can be before dd or dt elements inside dl elements.
dd can be after dt or dd elements inside dl elements.

that's the case in my example

I don't know if all UA do that but it's not the actual spec
http://www.w3.org/TR/html4/present/frames.html#edef-IFRAME
If all the UA do that, it's no enough to just say there is no need to author fallback, i think we need to say that this fallback content is given by the UA with something like IFRAME: "content of the src attribut on the iframe element"

Huh? <iframe> is just a way to embed another web page. Surely that page should be made accessible to the user. Not sure why you need fallback content for that.

I talk about fallback for non supporting iframe element UA (lynx for example)