Research suggests that pair programming increases student performance and decreases student attrition. However, less is known about the ways in which pair programming can unintentionally lead to inequitable relationships between students. Audio data were collected from pair programming interactions in a sixth-grade computer science enrichment program designed to promote equity. However, even in this context, there were surprising instances of inequity. We measured inequity by documenting the distribution of students' questions, commands, and total talk within four pairs. Analysis revealed that less equitable pairs sought to complete tasks quickly and this may have led to patterns of marginalization and domination. Notably, this focus on speed was not evident in the more equitable pairs. These findings are important for understanding mechanisms of inequity and designing equitable collaboration practices in computer science.