Large Language Models (LLMs) are increasingly used in advisory and collaborative roles in multilingual contexts where behavioral appropriateness matters alongside factual correctness. One common way to operationalize such behavioral tendencies is through personality-related assessment frameworks. However, it remains unclear how the personality-related response patterns generated by aligned LLMs compare with empirically observed human variability, particularly with respect to the breadth and structure of their response space. We examine personality-related response patterns of three contemporary LLMs using the HEXACO framework in four languages (English, German, Spanish, and Finnish). We contextualize facet-level response profiles using published human reference data for the corresponding language versions and analyze them using repeated controlled questionnaire-based elicitation. Our results show that human personality assessments exhibit structured cross-linguistic variation embedded within a relatively balanced response range, whereas LLM-generated Likert-scale responses cluster within a narrow, normatively positive region of the scale. The pronounced ceiling effects on socially desirable dimensions, such as fairness and prudence, persist across models and languages. Although LLMs display some language-dependent variation at the facet level, these shifts remain localized and lack the coordinated structure observed in human reference profiles.