Knowing the number of virtual machines (VMs) that a cloud physical hardware can (further) support is critical as it has implications on provisioning and hardware procurement. However, current methods for estimating the maximum number of VMs possible on a given hardware is usually the ratio of the specifications of a VM to the underlying cloud hardware's specifications. Such naive and linear estimation methods mostly yield impractical limits as to how many VMs the hardware can actually support. It was found that if we base on the naive division method, user experience on VMs at those limits would be severely degraded. In this paper, we demonstrate through experimental results, the significant gap between the limits derived using the estimation method mentioned above and the actual situation. We believe for a more practicable estimation of the limits of the underlying infrastructure, dominant workload of VMs should also be factored in.