Trade-off between validity and efficiency of merging p-values under arbitrary dependence