Online source rate control (RC) is designed for video streaming over high-speed packet access (HSPA) and Long-Term Evolution (LTE)-style variable bit rate (VBR) downlink channels. The problem is formulated as the adaptive adjustment of the operational mode of a video encoder based on the buffer overflow probability (BOP) feedback received from the radio link control (RLC) layer at the base station (BS). This allows us to maximize the attainable visual quality while keeping the transmitter BOP below a desired threshold and maintaining a video delay as low as possible. We derive an online measurement-based BOP estimation model for the RLC buffer, which is capable of operating with no prior knowledge of the channel variations and of the video characteristics. Based on this estimation model, an online adaptive RC algorithm is proposed to seamlessly adapt the bit stream to the characteristics of VBR channels. Our experiments are conducted in multiuser scenarios using VBR video encoding combined with adaptive modulation and coding (AMC) in the transceiver. The results demonstrate that the proposed source RC regime supports near-instantaneous yet smooth bit stream adaptability, which makes it useful for HSPA and LTE-style systems for the sake of accommodating unknown video traffic characteristics and dynamically fluctuating propagation conditions.