cgroup throttle on child device does not work

We're running Docker containers with devicemapper on CentOS 7 (3.10.0-514.10.2.el7.x86_64).
For some containers we can easily throttle reads through blkio.throttle.read_bps_device by specifying the parent block device, but for some containers it just doesn't work.

Here's our lsblk:

NAME                                  MAJ:MIN RM  SIZE RO TYPE MOUNTPOINT
sda                                     8:0    0   32G  0 disk
└─sda1                                  8:1    0   32G  0 part /
sdb                                     8:16   0  100G  0 disk
├─data-docker_thin_tmeta              253:0    0  512M  0 lvm
│ └─data-docker_thin                  253:2    0   90G  0 lvm
│   └─docker-8:1-151197496-c619...6ed 253:4    0   10G  0 dm
├─data-docker_thin_tdata              253:1    0   90G  0 lvm
│ └─data-docker_thin                  253:2    0   90G  0 lvm
│   └─docker-8:1-151197496-c619...6ed 253:4    0   10G  0 dm
└─data-data                           253:3    0    9G  0 lvm  /data

When we specify in a cgroup the following, it properly throttles the reads for some containers:
echo "8:16 1000000" > blkio.throttle.read_bps_device

But for other containers, we need to specify the exact device such as:
echo "253:4 1000000" > blkio.throttle.read_bps_device

Any ideas what may be the reason why the throttling with parent device at 8:16 does not work sometimes?