The VideoFrame tag is used to render one frame. It includes the data of exactly one video frame to be drawn on the screen.
The f_object_id_ref parameter is a reference to a DefineVideoStream.
The f_frame parameter defines which frame needs to be rendered. Note, however, that is not enough to display the video frame in the output. For that purpose you also need to use a PlaceObject2 or PlaceObject3 with their morph parameter (f_morph_position) set to the same frame number. This method limits the videos to 65536 frames (about 18 minutes of video). Longer videos can be created using multiple video stream blocks.
The f_video_data content depends on the codec defined in the DefineVideoStream tag. Once I really know what that data is, I will update this documentation. Note that the Sorenson H.263 encoding is actually a subset of MPEG-2.