That last test is amazing, i'm trying to wrap my mind around what is happening.
I can sort of see it,.
...the TC taking everything that the lens is throwing at it, and since it is "1.4x" it can't use the outermost part of an EF image anyways, as adding that 1.4x obviously means a reduced F.O.V.,. it's essentially magnifying a crop of an EF Lens image circle.. the crop happens to be about the size of the EF-S' projected image.
But since the TC is designed to work with full frame, the output image from the TC is essentially scaling the cropped image circle up.
On first thought, I'd never expect this to work. I can;t imagine gaining anything but poor IQ from adding a TC to such a wide lens.
But it looks like, almost coincidentally, that it just happens to be adding up!
Yeap, as its designed for a 1.6x crop and actually a 1.5x (nikon) the image circle is just too small on its own but using the TC its magnifying everthing by 1.4X and hence spreading the image circle over a greater area which in this case is enough to cover a FF sensor.
Sorry MOD's must be getting close to my limit but here are a couple more samples, this time agaist my 24-105L!
Full image (5D2 - 24mm 24-105)
10-20 +1.4TC @ 16mm
5D2 100% crops 24-105
5D2 100% crops 10-20+TC - Yes more sharpening is required and it is a higher ISO
Now resized to 40D size (10MP) 100%crop.
And finally a shot at 10mm+TC, the distortion is really not bad at all.
I really can't get over how good this comb is!