It appears that when cdc_ecm was refactored it changed from using tx_thread_sleep to ux_utility_delay_ms but one of the sleep constants being converted from ms to ticks was not adjusted with the refactoring.
ux_host_class_cdc_ecm_thread.c line 157:
_ux_utility_delay_ms(UX_MS_TO_TICK(UX_HOST_CLASS_CDC_ECM_PACKET_POOL_INSTANCE_WAIT));
The impact will depend on the system's UX_PERIODIC_RATE but I believe the default is 100 or 10ms per tick. In which case the sleep will be 1/10th as long as desired.