A solution mentionned above was to use CPLD. In this case, how do you communicate with the microcontroller without using mapped I/O. Sérial link ? Shift register ?
Using a CPLD, with the 668 internal, is not a very elegant solution. If I had to, I would memory map the cpld at ffxx and move what I had on P0, P2 and p3.6&7 to CPLD outputs.

Describe the number of bidirectionals, outputs and inputs you need and I'll think about it.

Erik