I can verify this. And it’s a oh so common problem that various OpenGL/ES implementation mess up the 4th component of texcoords (don’t ask me why - it’s just a homogenous coordinate. That stuff is not difficult to understand).
The emulator at least has correct behaviour in some cases. The implementations on the real PowerVR chips I’ve seen (a workmate works with them) do it wrong all the time, but to o be fair, all other OpenGL/ES implementations I’ve seen so far do it wrong as well.
Btw, using the texure matrix to inject non-one w-coordinates into the texture coordinates doesn’t work either. I think the 3rd and 4th component is silently ignored.
Imho proper handling of homogenous texture coordinates should be part of the OpenGL/ES tests.