The reason it works on traditional flat panels is that there is a one to one correspondence between the image data and the panel layout. That is not the case with PenTile displays and there many different subpixel arrangements so it is not possible to predict what subpixels on the display will lit up for a given subpixel in the image data. For best results you also need fonts designed for subpixel rendering so even if subpixel rendering was practical on PenTile displays you would carefully need to tune the fonts for every possible PenTile subpixel arrangement.
Well, getting the hardware (manufacturers) to tell you what the hell it (their stuff) is doing is always a problem and even extremely complicated designs and massive marketing efforts like USB only kind of solve the problem for the most common of the common cases, but that’s kind of a truism.
Re you second point, I originally wanted to say something like “fonts aren’t, renderers are”, but in the meantime I read the Raster Tragedy[1] and it seems like the answer to your question is quite literally yes, hinting bytecode is in fact designed for a subpixel grid with a 1:3 aspect ratio (mostly by virtue of the font designer writing and checking it by feel using one). Shoehorning outline fonts and WYSIWYG layout onto low-resolution (≤ 150 dpi or so at 10 pt) displays is that much of a hack, and now I really, really want it to die, all the complaints about elitist designers notwithstanding. (That I find 200- and 300-dpi tablets and phones much more pleasant to work on is a mere coincidence, I assure you.)