There is a logic bug in the Terracotta JobStore (DefaultClusteredJobStore class, to be precise).
It also appears that the same bug likely exists in RAMJobStore.
Furthermore I can see that there is a semantic problem with the JobStore.pauseTriggers method that leads to this problem, and the bug may likely also exist in JDBCJobStore as well.
The problem is that the method now (since 2.0) takes a GroupMatcher, which isn't necessarily an exact group name (e.g. could be a start with or contains etc. rather than an equals). That means the current set of groups must be iterated over to see which ones match. If there are no triggers in the group, then it isn't there to be iterated over and considered a match!