Also to get more DMA cycles, you can set the width to 32 characters and turn off doublewide. It makes your horizontal resolution smaller but will get you more cycles.
You can also use 320D mode. It costs less DMA time (like 320A) but has the limitation of every odd numbered column having different colors from the even numbered ones. You also have to turn on Kangaroo mode for it to work, but it may be possible to change the colors more reliably with interrupts. 320D has no normal practical use, but might work in this instance.