I don't think caches speed up a streaming read or write operation. After all, the test is supposed to show reading from RAM, not from the CPU caches. One may safely assume that the latter are fast.
EDIT: the apollo core can read 8 bytes per cycle from its dcache and write 8 bytes per cycle to the dcache (at the same time!) resulting in something like 700 to 800 MB/s each for read and write or a total throughput of one and a half gigabytes per second. Not bad for a 90 MHz processor, huh?