I have not considered direct SPI as I see no circumstance where read operations won't benefit, as the Arduino does not have enough ram for a screen buffer. But this mode can be inserted easily.
There is only one issue with the Atmel HW SPI subsystem. Of course we need clock and data out lines. But additionally the HW SPI subsystem occupies the "data in" line. So we loose one line here (at least to what i know, and i never was able to use this pin for something else)
It takes nearly one second to repaint the entire screen.