This paper investigates the potential of improving the Quality of Experience (QoE) of pervasive video delivery by buffering. It reveals the importance of delivering video without freezes to the end user. It then introduces a fluid flow model that leads to a closed formula for the freeze probability, and matches measured distributions of on- and off-times to this model's exponential distributions in order to underline its feasibility. The closed formula for the freeze probability is then used to investigate the impact of two buffer size adaptation policies, one additive and one multiplicative with regards to the deviation of the average off-time from its nominal value. It is proven that, besides of some specific conditions in the additive case, the freeze probability grows with the disturbance, which means that even an increased buffer size implies worse performance and thus QoE. It also points out a way out of this dilemma, which is to try to reduce off times by reducing the utilization of the mobile link.