Lennard: I got your email but reply here instead.. Interesting findings.. A couple of questions:
(1) Do you call GXEndDraw after you have copied all data using 32-bit DWORD writes?
(2) Is the display double buffered? I.e. if you do not call GXEndDraw and fill only half of the display with some color, is it updated on the display?
(3) If you call GXBeginDraw before filling the same amount of system memory, is the performance the same? (The device could have a thread running that simply copies the contents of the RAM buffer to the display - providing support for those applications that only call GXBeginDraw once).