The rapid development of in-network computing motivates people to consider deploying virtual network functions (vNFs) on heterogeneous platforms that include both software systems like virtual machines and docker containers and hardware systems like programmable data plane switches. Meanwhile, with the emergence of multi-client network services, service providers need to build vNF service trees (vNF-STs) in their substrate networks (SNTs). In this work, we study how to optimize the management reconfiguration of vNF-STs in an SNT equipped with heterogeneous platforms. An integer linear programming (ILP) model is first formulated to consider three common reasons for vNF-ST reconfiguration and reduce both the total resource usage after reconfiguration and the overall vNF migration cost during reconfiguration. Then, we design a two-step algorithm to reduce the time complexity of problem-solving. Specifically, the algorithm first checks all the active vNF-STs to select the vNF-STs that should be reconfigured, and then leverages an approach based on layered auxiliary graphs (LAGs) to reconfigure the selected vNF-STs. Extensive simulations verify the effectiveness of our algorithms on optimizing the management reconfiguration, and demonstrate that they can outperform existing benchmarks.