You should team up with the NTSC video out project, since you will need a similar frame buffer.

For the purpose of learning about frame buffers, look into these chips:

ac157 for address multiplexing
hm6264 or equiv - 8k x8 static ram
act715 ? maybe to generate VS and HS - probably overkill for an 8x8 LED matrix,
but so is a frame buffer.

"DMA" - it would be a shame for the 8051 to not have "DMA" to its own frame buffer. "DMA" applies to peripherals. The 8051 should be able to write/read directly into the frame buffer. This can easily be done by alternating access to the memory.... 8051 on even cycles, shift register on odd cycles.

You need some counters to generate an address to read out the memory. The data out goes into a shift register, such as the one erik mentioned.

I don't think using 8 of those to support an 8x8 matrix is a good idea. Those chips are designed to sink lots of current, more current than a SINGLE led can handle continuously. You would have to have a resistor in series with each LED. When you matrix(multiplex) them, you increase the peak current to the LED by 8, but the avg current stays the same. You only need 8 resistors for the matrix, vs 64 for direct drive.

The don lancaster "cheap video cookbook" should be required reading.

It's perfectly valid, if not learning frame buffers, to make the 8051 do the multiplexing, with no external frame buffer.