Typically it is < 1ms, when the network and the number of client requests per second is not taken into consideration.
The network delay can differ very much, depending on the traffic load. That’s why NTP packets contain two time stamps (packet output and packet input). So the response time can be calculated and compensated.
For standard NTP, the network delay (response time) is not relevant for the precision of the client synchronization.
Except, when the delay should be very high, then the reachability of the NTP server could be degraded.