Linear terminal control laws in model predictive control imply the need to compromise between performance and size of terminal region. A remedy allows the control law to vary with time, but requires very large online computation. Here, use is made of nonlinear terminal laws comprising an interpolation between predefined laws, each with its own precomputed terminal set. This reduces the online computation significantly, yet results in large terminal sets defined by the convex hull of invariant ellipsoids. The use of a different prediction class also produces significant benefits with respect to performance.