IF all the lines were exactly balanced in phase length, then yes, the phase shifter would be exactly set at 90 degrees. In practice, though, you find that the line lengths are not balanced. For instance, maybe you need more power on one port of the mixer/phase detector than the other input port. So you add an amplifier to that port. That, of course, screws up your phase balance.
So you temporarily hook a DC voltmeter to the phase detector/mixer output, and adjust the phase trim until you read 0 volts DC. Then you remove the voltmeter (so it does not add noise) and proceed to make the phase noise measurement. A mixer acts as a phase detector when it is tuned up in "quadrature", which means its IF port output is at 0 VDC, with small ac variations (that contain the phase information). If you have an automated phase noise measurement system, it will do these adjustments for you as part of its narrowband phaselocking system.
The reason this all works (being able to use a relatively poor microwave signal source for a high dynamic range phase noise measurement) is that you are measuring the phase noise variations of the dividers themselves, but the phase noise of the microwave source is "correlated" at both divider outputs, so its inital phase noise "cancels out".
IF your measurement system does not set the phase dectector's IF output to 0VDC, then you will have less phase detector gain AND you will also be measuring AM and PM noise combined.
Rich