Game development has evolved from various software paradigms into component-driven architectures, such as those used in Unity 3D and Unreal Engine. The recent rise of generative AI (Gen AI) and large language models (LLMs) introduces new opportunities and challenges for content generation in games, particularly regarding real-time integration. This paper investigates the performance implications of incorporating real-time LLM-generated content into two popular Unity 3D architectural patterns: Object-Oriented Programming (OOP) and Data-Oriented Design (DOD) via the Entity Component System (ECS). Using a controlled prototype, we compare the performance of these architectures under increasing loads of real-time and pre-generated LLM content. Our findings provide practical insights for game developers considering Gen AI integration and highlight the architectural trade-offs involved. We observe that ECS outperforms OOP under high-frequency LLM query loads.