One of the problems you may have is with DV25 4:1:1 sampling. You can't really avoid that if your final output is to DV. Every 4th pixel has a definite red and blue value but the three pixels in between must be interpolated.
It might be that if your text is white (235) it'll look better since every pixel has a luma value. Bigger certainly helps, as do strong verticals instead of diagonals.
Preview it on a monitor or TV. You may find that the fuzziness just fades into the general smear of TV.
Create them large and resize down with pan/crop? yes. Even define a track for text that is 2x 720x480 for zooming in/out. if text is moving at high rates, I increase the number of frames per sec on that track, this helps with smoothness and permits frame by frame inspection for distortion.