I used a similar to Bruce's approach, originaly designed for a 8x8 matrix array, but adapted for 7 segment multiplexing. It used one port for data transfer, two latches and 2 lines for latch selection. It can be used to multiplex up to 8 digits and allows any pattern suitable for 7 segment display to be displayed.
I also used 1 timer to syncronize which digit will be active. Every 5ms, the timer interrupts the cpu, checks what digit to display, blanks display, select digit, latch it, output the data and then latch it. This refreshed the whole display 50 times per second, so it is completely viewable by the human eye as a continous display, and uses only about 2% of cpu time.