the use of dma can be good there...
i've read somewhere that dma transfert can be done between two memory areas.
dma is like a pre-caching function...
or, in the case of sb, it is a data flow.
the µP can connect where he want's in the current dma transfert, where, whose dma channel, how many,

how to program the dma chip?
can be good for memory and disk copy.
the screen refresh for exemple, is it possible with dma?

PCI devices are able to bus master and read directly from memory, iirc (I haven't begun to code in that direction just yet). Getting the responsiveness from modern hardware certainly requires orchastrating many things going on at once - not just what the processor/s is/are doing.

It is a daunting task with many mistakes to be made.

[Currently, I'm working on a compression algorithm - the decompressor is only a couple bytes, so it could be used on a boot sector or second stage loader with great success. Now if I could only compress data without brute force it would be more useful...]

compression to make faster caching??

why not!