If you understand how the RPC works, you will understand why the capacitor method is unstable, and why using a transformer is in some ways better. It is not ideal, the ideal way is a change in the idler design. Also, the phase angles will be very close to perfectly at 120 deg, unless mucked with using capacitors.
1) A three phase motor (or single phase, etc) produces a "back EMF", a voltage that opposes the applied voltage. That is an essential part of how it works. That voltage is what cuts the starting "inrush" current down to the load current required by the shaft power output needed. Each phase of that has to exactly oppose the incoming voltage, so yes, it WILL be closely at 120 degrees, and improves with load..
2) The back EMF is ALWAYS less than the effective line voltage, because the difference between them is what drives current into the motor. The difference will vary with load, but it exists.
3) The "generated leg" is actually the back EMF if the 3rd phase wire. Remembering that it MUST BE less than the effective mains (line) voltage, you can easily see that it will need to be boosted. The idler motor can be designed with an intentionally higher back EMF on the generated leg (more turns of wire), or it can be boosted with a capacitive compensator, or it can be boosted with a transformer. You can even ignore the difference, but it may require derating load motors.
The best way is the designed-in extra turns. The transformer is also good, although it adds impedance in the generated leg, resulting in more voltage drop, but it is absolutely predictable as far as output voltage. The capacitor method is the most popular, but the least stable, because it is very loading-dependent, and because the light load performance need causes the heavier load performance to be compromised, you cannot tolerate the high voltage that would occur at light loads if the heavy load were compensated fully.