But you definitely can't take a "regular driver" and change it quickly for this.
Show me a typical 3-wire hc164 solution and I can make limited changes to get to run on 2 wires and then 1wire.
The beauty of this is that the final implementation
using 1 wire is actually faster than the LiquidCrystal library using 4 bit mode.
That will depend on how many wires you want to reduce. The growth in duration is exponential with the number of pins to be reduced, and it is actually quite difficult to implement in applications where other interrupts are going on.
But putting aside LiquidCrystal, I can update a 16-char line on a less than 2ms, using a shift register and on a 1MIPS avr.