Metaverse technology is becoming increasingly popular in educational environments around the world. However, before it is widely applied in public education for adolescents learning English as a foreign language (EFL), it is crucial to assess its effectiveness in comparison to traditional teaching approaches. Our study involved 336 seventh-grade students who were recruited from three public middle schools and divided into two experimental groups (metaverse-mediated and face-to-face speaking) and one control group (textbook-based class). After a 12-week intervention, we assessed improvements in English speaking, listening, vocabulary, and attitudes, and we used questionnaires and teacher interviews to obtain qualitative feedback. The results indicated that face-to-face instruction was superior for improving speaking skills, providing better individualized support within students' zones of proximal development compared with metaverse-mediated and textbook-based methods. Metaverse classes, however, proved beneficial in reducing peer pressure and increasing engagement among students who were self-conscious about their English proficiency, due to the non-direct interaction and immersive, gamified environments. In terms of listening skills, the textbook approach outperformed both the metaverse and face-to-face methods, likely due to its focus on listening activities. Vocabulary improvement was similar among the groups, suggesting that the instruction modality (virtual or in-person) was less important than adequate language exposure. These findings highlight the need for careful consideration of the metaverse's strengths and limitations before integrating it into EFL public education.