DVMs do have limits to compensating 4-wire connections, there is usually a specification called out in the manual for this and whatever effects it may have on the accuracy, DVMs measure and mathematically subtract out lead errors but it is not entirely accurate. It still depends on the DVM's accuracy of measurement which is limited at such low values. The Kelvin bridge was developed to MINIMIZE errors due to interconnecting lead resistance, it does not eliminate the effects. In more modern bridges such as the ESI 242D, there are two adjustments which compensate for lead and yoke resistance up to 0.1 ohms by modifying resistance in the bridge arms to compensate for the extra resistance, these adjustments come very close to eliminating errors within the limits. There is also a 6-wire modification of the Kelvin bridge which further compensates for other errors at very low resistances, the NIST website has information on the various methods of resistance measurement.
No measurement technique is entirely free of error, each one has its own sources of error and knowing just what those errors are can help in making more accurate measurements.