On the picture below you can see that I have had problems with ghosting, and it annoyed me a lot.

So, on the right side you have my rev. 2 prototype (second PCB layout, not counting the early bread board prototype). It is usable, i.e. one can read the LCD, but there is also significant ghosting which makes the display unreadable when viewed from angle. I tried to modify the firmware (BTW firmware is based mostly on this http://www.dataweek.co.za/article.aspx?pklarticleid=2382), but couldn't do better than this. Since I didn't know of any simple method of generating 1/3 bias wave forms on bare GPIOs, I decided to try out the cheapest driver I could find on farnell (that was PCF85176 at the time of writing : http://www.farnell.com/datasheets/1923814.pdf). And it performs much better, as you can see on the left side of the picture. Another cool thing is that now µC uses only 2 pins for driving the whole LCD, since the PCF85176 communicates via I2C instead of 16 pins which were used to drive the LCD directly. This fact will allow me to use smaller micro (STM32F042 or 72 or 70) which should be much cheaper than the one I use now. I expect that overall cost should be the same or even lower.