A magnetic levitation plant is a fast, unstable, nonlinear mechatronic system vulnerable to measurement noise and external interference. Designing a high-performance and robust controller for such system, immune to noise and disturbances, and ensuring a wide operating region, is a challenging task. One can chose a well-know LQI controller, designed based on a linearized plant model, to stabilise the system and to eliminate the steady-state error. However, large plant-model mismatch and considerable non-linearity of the system may limit the control quality provided by such a solution. To improve the system performance one can introduce some additional degrees of freedom by replacing the integer order integrator with a fractional one. For practical implementation, the latter may be approximated with an integer-order realisation obtained with the time-domain Oustaloup method. The paper explains the idea of an $\text{LQI}^{\beta}$ controller, presents selected results of simulation tests of the closed-loop control system behaviour, and provides discussion of possible applications and advantages of the proposed approach.