Process discovery studies algorithms that, given an event log of a system captured as a collection of recorded sequences of actions executed by the system, construct process models that describe the system. A process discovery problem is a multi-objective optimization problem that aims to simultaneously optimize the simplicity of the constructed models and their quality, ensuring models accurately represent both the input event log and the underlying system. Multi-objective metaheuristics provide effective strategies for navigating these trade-offs, offering practical approaches to exploring Pareto-optimal solutions. Recent research has demonstrated that genetic strategies based on grammatical inference can produce superior models compared to state-of-the-art discovery algorithms. In this paper, we conduct a comprehensive evaluation of existing multi-objective metaheuristics for process discovery using grammatical inference, refining them where necessary to enhance efficiency and control the number of the discovered Pareto-optimal models. This evaluation, based on multiple real-life event logs and our open-source implementation of various metaheuristic algorithms, confirms the feasibility of efficiently discovering significantly better models. Specifically, the Differential Evolution-based optimization approach, which we refer to as ADESPD, consistently produces high-quality models with diverse characteristics within practical time constraints, which are deterministic and sound.