In this paper we present strategies to deal with inherent load uncertainties in future generation mobile networks. We address the interplay between user differentiation and resource allocation, and specifically the problem of CPU load control in a radio network controller (RNC). The algorithms we present distinguish between two types of uncertainty: The resource needs for arriving requests and the variation over time with respect to user service policies. We use feedback mechanisms inspired by automatic control techniques for the first type, and policy-dependent deterministic algorithms for the second type. We test alternative strategies in overload situations. One approach combines feedback control with a pool allocation mechanism, and a second is based on a rejection-ratio-minimising algorithm together with a state estimator. A simulation environment and traffic models for users with voice, mail, SMS and web browsing sessions were built for the purpose of evaluation of the above strategies. Our load control architectures were tested in comparison with an existing algorithm based on the leaky bucket principle. The studies show a superior behaviour with respect to load control in presence of relative user priorities, and minimal rejection criteria. Moreover, our architecture can be tuned to future developments with respect to user differentiation policy.