What about internally rendering it at double the pixel dimension, then applying bicubic resampling before screen render? How about an OpenGL solution for hardware antialiasing support? Just a couple of ideas.