You original statement, that PLLs ALWAYS have low corner frequencies is wrong. I have done standard loops with up to 1.5 MHz bandwidth, and an injection locked loop is typically tens of MHz wide locking bandwidth.
If you do a careful noise analysis, you will find that for a microwave PLL that uses a low frequency crystal oscillator (below 50 MHz), then the optimum phase noise requires the control loop open loop bandwidth be somewhere in the 3KHz to 250 KHz range. Finer step sizes typically required smaller control loop bandwidths for both noise and spurious reasons.
But if you were locking a wideband microwave VCO to, perhaps a 633 MHz saw oscillator, and tuning in a 633 MHz step size, you would probably get optimum phase noise with a loop bandwidth somewhere in the 1 to 2 Mhz range.
Higher bandwidth loops require special techniques to keep some phase margin.