The best way to do the similar thing with a cheap ARM would be to do dual-writes without any writes.

Basically:

dummy = my_start_address + flag;
port = dummy++; // Output an address and clock
port = 0; // deactivate clock
port = dummy++; // output next address and clock
port = 0;
...

where port is a wide enough port that it will present an incrementinga address, and have one extra bit that gets set and cleared by every second write to clock some external logic.
This should take 4 clocks/cycle and be 20% faster than if the data is first read in and then written out. A 72MHz NXP LPC2364 should then manage 55ns/cycle. The internal PLL may then be used together with a suitable crystal to fine-tune the cycle speed - in your case you would need the ARM to run at 66.67 MHz.