Why do each gluster distribute set have different space utilization?

Issue

In distributed volume type there can be difference in space utilization between two sets. Example:

------ Connecting to node gluster-1 ------
/dev/mapper/glustervg-glusterlv   25T   22T  2.4T  91% /brick

------ Connecting to node gluster-2 ------
/dev/mapper/glustervg-glusterlv   25T   14T   11T  55% /brick

This applies to disctributed-replicated volumes as well - each replica brick in distribute set is expected to have same amount of data.