Problem was most of the demos used many many megs of ram as they procedurally generated objects and stored them in memory. The 64k demos were disk size.

Assuming memory mapping is used for displays the (128k of ram (64k of words)) is also needed to hold your display buffer. Assuming a mere 320 x 200 x 8 screen will use 1/2 of your entire virtual memory space. Look for demos of what people do on C64 demos for a better idea. Yes you can run a UNIX, yes you can talk TCP... but its going to chew up a lot of the capacity for that infrastructure. I expect very light weight protocols, TCP/IP is likely too heavy weight.