Preventive Maintenance activities (PMs) are essential for semiconductor manufacturing equipment. Since PMs are typically non-preemptive, they are usually modeled as high priority customers. We examine the accuracy of a popular approximation for the mean cycle time in G/G/1 queues with high priority PM customers. This approximation follows Kingman's and is based on a heavy traffic assumption. We observe that the approximations can perform very poorly when the mean and coefficient of variation of the PM durations are large and small, respectively. This is precisely the practical case for PMs in semiconductor manufacturing. We propose two practical correction candidates and assess their effectiveness. SPECIAL SESSION CODE = 4yd3b.