Peak always shows the high energy level or in other words One or more poles are dominating the frequency response.
Try to add any zero in the feed forward path by adding any buffer or by resistance in such a way it occures exactly at the dominant pole.
As the peaks occur at relatively high frequencies (100 MHz region) it seems to be a result of parasitic capacitive feedbacks. But - is it a problem for you resp. for the proper function of the system?
(By the way: it is not a "phase margin" plot - it`s simply a plot of the phase response).
It is typically means there are 2 parasitic poles close to each other (unless you managed to create a zero somehow). Think of an LC oscillator. You will see large Q and peak in the magnitude plot at the resonant frequency. Also you'll see 180 degree phase shift.