Parallel addressing will need 9 address lines, 1 data line, a latch signal and at least 30 discrete ICs (28 x 16 bit ports plus some glue logic).
Serial addressing would need 1 address line, 1 data line, a latch signal and shift registers
You might consider that PIC's make good shift registers themselves and have higher pin current capability than standard CMOS devices. Alternatively, the extra 'intelligence' they offer over simple shift registers means you might be able to develop a simple protocol to control outputs without daisy chaining them and it would be expandable too. For example you might use a master UART to address lots of PIC in parallel and send "449,1" to turn output 449 on and "449,0" to turn it off.
Brian.