Minimum variance (MV) benchmark is useful for identifying variance reduction opportunities in industrial control systems. During the past two decades, MV benchmarks for single-input single-output (SISO) and multi-input multi-output (MIMO) systems have been proposed. These MV benchmarks do not account for the structure of the decentralized or multi-loop controllers, which are used almost exclusively for regulation purposes in process industries. Due to this drawback, the available MV benchmarks can lead to incorrect conclusions regarding the performance of decentralized controllers. This paper aims to fill this gap. For performance assessment of decentralized controllers on a loop-by-loop basis, we present a simple modification of the available MV benchmark for SISO systems. For simultaneous performance assessment of all loops, we present a method for computing a tight lower bound on the achievable output variance. In the latter approach, the non-convexity of the resulting optimization problem is handled using sums of squares programming. The usefulness of the proposed benchmarks is evaluated using examples drawn from the literature.