Theoretically you should be able to read the VBI data directly from the MJPEG frame. This would be problematic with MPEG however. I would recommend capturing at least 1440 px per scanline. The vertical doesn't matter as long as you are getting all 525 lines (526 stored lines since one line is displayed as two half-lines)
We'd probably want a lossless frame compression too. MJPEG wasn't designed for the discontinuities found in the VBI. In other circumstances I actually wouldn't mind for lossy for video.