Most PCs have one memory bank accessed by all processors, so if all you want is to have both processes see the same data then shared memory is your best bet.
I believe that the preferred way to use shared memory is to mount a ram disk on the hard drive (shmfs or tmpfs), and to have both processes read and write files on that directory (which physically exist in system RAM). That results in understandable and maintainable code, and can be easily ported to other systems (including mainframes, which may have much more complex memory arrangements).
But if minimum CPU is the goal then you may want to skip the virtual filesystem and access shared memory segments directly; there's quite a good tutorial on this here:
http://www.cs.cf.ac.uk/Dave/C/node27.html. This is much more code to write than the filesystem approach, but has the advantage that there is no reading or writing to do; the data is simply there for both programs whenever it exists.
I should also note that the most "correct" way to share information is to use a pipe (named or anonymous). This has the advantage that you can't run out of memory: if the process sending information sends too much for the OS to buffer, then it'll simply slow down the sending process until the receiver has processed more. This is even simpler than the RAM disk approach, as it needs no setup other than that which the program can do.
The Documentation Project has further reading on IPC:
http://tldp.org/LDP/lpg/node7.html
Hope that helps,
—Robert J Lee
