I, personally, most often use memory mapped I/O to achieve the number of pins I need. I do, however, resort to IIC when I only need a few more pins than I have. The software is available in a million appnotes. Another advantage is the 2 pins will do however much you hang on. Re the IIC as opposed to shiftregisters in this application is that the IIC parts can generate an interrupt.
I do like your 4 pin (in / out / shift / latch) solution for the economy and in this case it may be the right way.
TGIF
Erik