Shadow and nested paging are two means to enable the memory access for a VM guest. Although the nested paging can eliminate the need of vmexit due to frequent guest page faults, new form of overhead is meanwhile introduced by the two dimensional page table walking. Even TLB is employed as a quick reference in the address translation, the performance gained may in a certain cases still be offset by the new overhead. The fact is, each method is only capable for handling a subset of the whole workloads, neither approach is guaranteed to win the race and deliver the best performance for all types of the workloads. A possible attempt is to enable the VMM to reconfigure its paging method at run-time as a response to the ever changing behavior of the workloads in memory accessing. We describe the design of this mechanism, and part of the implementation based on the QEMU-KVM virtual machine monitor.