I was just looking over Ace's web page. I don't recall anyone else here mentioning this, but I think it is relevant. (At least, relevant to type of analysis. Ace's interpretation is farking nuts anyway.)
I see no mention of deinterlacing in Ace's work. De-interlacing an NTSC (or PAL) video signal to display on progressive scan screen can cause the edge of the object to wiggle - depending on the de-interlace method used.
If you just mash the two 240 line fields into one frame, then you will get the most extreme problem. The two fields are taken 1/60 of second apart. Moving objects will therefore show position differences between the two fields. This will make the edges of the object look like the teeth of a comb. The faster the object moves, the longer the "teeth" will be.
At Ace's measured 20 pixels per frame for the airplane, that's going to give you "teeth" ten pixels long. When you then compress the video, those "teeth" will be smeared together. You will in effect have an indistinct outline of the plane, which may be anywhere up to 10 pixels from where it appears to be.
If a more sophisticated de-interlace method is used, things don't really improve for Ace. The more sophisticated methods eliminate the comb teeth, but they tend to move the objects a few pixels. Depending on just how each field pair lines up, the de-intelacer may decide to align the image to first the even field or to first the odd field.
In one frame, your object may align to the odd fields. The next frame may align to the even field. In this case, you would get a significant apparent slow down, namely a drop of half of the measured travel distance between the two frames. The other way is just as bad. If the de-interlacer aligns in one frame to the even and in the next frame to the odd, you will get an apparent speed up of the same magnitude.
So, de-interlacing in whatever form is going to be a problem for measurements. And that's ignoring the problems you can get into if the de-interlacer is started one field off. That is, if your video starts on an even frame and the de-interlacer doesn't compensate.
Even worse, the comments others have made about video compression are quite correct. MPEG compressed video stores only some frames completely. Others are stored as a sort of mathematical description of the differences. Typically, you might have one complete frame followed by four that will be calculated.
Part of that "mathematical description" is an approximation of the motion of the objects visible. This motion compensation is a further source of error in Ace's calculations that he has not bothered to account for.
The motion compensation in the compressed video will not be perfect. The objects don't all move at the same speed, and there is (as always) a tradeoff of accuracy against computation power. More accuracy takes more time for compression, and more space in the finished file. In heavily compressed videos from YouTube, the number of calculated frames will be higher, thus each frame will have even more potential for error.
It is, therefore, quite reasonable that only the full frames in the compressed video will show object in their correct positions. Reasonable and expected, except by Ace.
Anything else? Well, the thousands of witnesses, but Ace doesn't count them. He can't line them up on his monitor and make them lie the way he can torture a video into telling his Truth.