Embodiments of the present invention provide techniques for efficiently coding video data during circumstances of network congestion, or instances when a decoder is unable to handle incoming video at the intended frame rate. A video coder may code frames of a video sequence according to motion estimation prediction in which each frame of the video sequence is constrained to be coded as one of an I-frame or a P-frame. The video coder may determine for each coded P-frame, a distance from the respective P-frame and a most-recently coded reference frame, and if the distance matches a predetermined threshold distance, the respective P-frame may be marked as a reference frame. The coded video data of the I-frame(s), the reference P-frames and the non-reference P-frames may be transmitted in a channel.