The server is back up. I constructed a tiny benchmark to compare speeds of the different interfaces. I wrote 80 characters, then 80 blanks, looped through it 20 times. Of course it goes too fast to read, but here is what I measured:
8 data pins + RW 432 milliseconds
8 data pins - RW 641 milliseconds
4 data pins + RW 719 milliseconds
4 data pins - RW 1038 milliseconds
The options with RW are testing the LCD busy flag, of course, so testing the flag is about 1/3 faster. Again all of the options go faster than you can read. And if complex calculations were interspersed with sending the characters to the LCD you would see even bigger speed increases.