Currently if a large amount of data is deleted simultaneously, the master will not do any throttling of the DeleteTablet requests send to the tservers. The tservers will use up to the configured number of service threads to work on deleting tablets. The deletion can be relatively heavy-weight – lots of file system operations to hole-punch out the dead tablets, etc. This can have a negative impact on other concurrent workloads.
It would be desirable to do some throttling either on the master or tserver side to avoid overwhelming disks and thread resources during heavy deletion.