> I'm guessing it is just some funny facet of the algebra concerned. The (normalized) transfer function for 'stage n', n=1 to 8 (so 30dB is n=5), appears to be
>
> H(s)=1/((s+1)^n+k*(s+1)^(2n-8))
>
> where k is the gain in the feedback loop. If one were to be bothered to work out the phase from this, tan^-1 im(H(wj))/re(H(wj)), I dare say it might be possible to see what it is about n=5 that keeps it nearly zero for w<1 (for 'w', read 'omega'). I have to say I'm not even remotely tempted, far too busy doing other things!

this is the best post i've ever read, may i put it on a t-shirt