Only if you assume there will be no checks that nodes do have data (which is what a GET is, that is the quality it brings - for popular data), even though it has been said in this topic that that is to be implemented.
Paying on upload is not necessarily that temporary. It is in fact simpler than pay on download, and pay on download does not provide anything unique that motivates the added complexity. In fact, as described in a post above, it introduces problems with encouraging false gets, without guaranteeing permanence of all data.