Speaking two or more languages shows remarkable bilingual flexibility. Yet flexible switching between languages often incur performance costs, suggesting competition among the currently activated languages. Here, we investigated language competition using three languages (L1, L2, L3; German, English, French), taking n-2 repetition costs as a marker for persisting inhibition. These costs denote the finding that performance is worse in n-2 repetition sequences (e.g., ABA) than in n-2 non-repetitions (CBA), suggesting persisting inhibition of switched-away languages. Two experiments tested if inhibition is influenced by the relative dominance of the currently competing language. Experiment 1 (n = 28) showed n-2 repetition costs only for L2 (English). Looking into L2 trials specifically, we found n-2 repetition costs only when switching back to L2 from the still weaker L3 (French) but not when returning from the stronger L1 (German). Experiment 2 successfully replicated this specific pattern with a larger sample (n = 44). Because L1 produced the best performance overall, consistent with always being the relatively most dominant language, finding no specific n-2 repetition costs for L1 may be due to strong general inhibition of L1 in all trials (i.e., L1 would get inhibited even when switching between the two non-dominant languages, so that L3-L2-L1 sequences would not lead to weaker L1 inhibition than L1-L2-L1 sequences). Yet, the focused analysis of L2 trials is consistent with the theoretically assumed relation between language dominance and language inhibition, suggesting that L2 is a strong competitor (requiring inhibition) for L3 but less so for L1.