Looking at this datasheet it looks like all commands except clear executes in 72us which is like 1/12 of the 900us you're telling the compiler it takes. Obviously you'd need to add additional delays to any CLEAR SCREEN command, like
Code:
LCDOUT $FE, 1
PAUSEUS 1500   ' Extra delay for this specific command which takes 1600us instead of 72us like all other commands.
If the two "command statements" in the inner loop took 72us each instead of 900us each the the whole thing would take 5.2ms instead of 144ms.