We investigate flow-level stability of schedulers in parallel-service wireless systems, which is important for maximizing the base station's capacity to serve the heterogeneous flows that are within the base station's power range. We model such a system as a multi-class queueing system with multiple preemptive servers, in which flows of different classes randomly arrive and depart once their flow is completed. The channel condition of a flow varies randomly over time because of fading and mobility. The evolution of the channel condition is assumed to be Markovian and class dependent. We focus on a general family of the best-rate schedulers that, whenever possible, serve flows that are in the channel condition corresponding to the highest achievable class-dependent transmission rate (i.e. the best rate). We prove under mild assumptions that any best-rate scheduler achieves maximal stability, that is, stabilizes the system whenever possible, in all systems with generally distributed class-dependent arrivals and flow sizes.