If I recall correctly this problem can best be approached from the Nyquist criterion for closed loop stability and the Principle of the Argument from complex numbers. It has to do with the number of encirclements of the critical point (-1) when doing a transformation from frequency to the complex domain (Re(H(jw)) vs Im(H(jw)).
Since a 180 degree phase shift from the positive Real axis lands you on the negative Real axis you have to watch for encriclements around -1. This will occur if you phase is greater than 180 at the 0dB (it implies a crossing of the Real axis to the right of -1). However, if you have a system that has a phase that goes beyond 180 degrees but comes back before 0dB then you are not encircling the critical point but rather moving past the Real axis and then coming back again to the left of the critical point.
So you're system remains stable. I hope this makes sense. It's tricky to describe without a picture.
Hope this helps.
Nino
Added after 1 minutes:
Oh yeah, and one more consequence is on your gain margin. If your gain isn't as you predicted in simulation you may have stability problems if it drops low enough to cause that dip in your phase plot to actually occur at 0dB.