What the system sees with a "standard" squirrel cage induction motor connected directly to the line is the inductive loading caused by the nature of the magnetic circuits within the machine design. This can result in a reactive power demand that is anywhere from 10 to 30 percent of the real power rating of the machine (e.g. power factor lies between 0.900 and 0.700). It also gets worse (i.e. farther from unity power factor) as the machine speeds up and/or becomes unloaded.
The utility sees a nearly uniform power factor when the variable frequency drive (VFD) is in operation; at least, for drives that have active front ends (AFEs). The power factor CAN get a little worse as the output voltage and frequency are phased back to provide slow speed operation of the motor, but the difference between full load and no load is usually no more than about 5 percent (roughly from 0.985 PF down to 0.940 PF). As you can see - even in the worst case, it still appears to be a "better" load from the utility's power factor measurement.
Adding power factor correction capacitors (PFCCs) between VFD and motor is not advisable since it really messes up the ability of the drive to adequately and accurately control the motor performance. In fact, there should be NO capacitive elements (including lightning arresters and surge suppression devices) in the circuit between VFD and motor.
From the facility viewpoint, the existing PFCCs may be superfluous, since the overall system power factor is going to come up toward unity as the VFDs are added between the motors and the line. (NOTE - this only applies if the VFDs REMAIN in service between motor and line, which means the VFD is not used as a "starter" and then bypassed when the motor is up to running speed!)