This is expressed as a percentage between 0.0 and 1.0. Between the low and high threshold, throttling is carried out by introducing a per I/O delay, which is linearly increased from 0 to  filestore_queue_high_delay_multiple/filestore_expected_throughput_ops.

And then from the high threshold to max, it will throttle at the rate determined by filestore_queue_max_delay_multiple/filestore_expected_throughput_ops

Both of these throttle rates use the configured one, which is expected throughout of the disk to calculate the correct delay to introduce. The delay_multiple variables are there to allow an increase of this delay if the queue goes over the high threshold.