The horizontal line is not a problem: the system has one more pole than zero, so there ought to be an asymptote. The grouping near z=1 is indeed confusing, but with little hope for improvement. Plotting the root locus with a very dense grid (see code above) reveals that there is a pole of multiplicity 6 near z=1. As a result, the computed roots in a ball of radius eps^(1/6) = 2e-3 are essentially random, which makes any pairing difficult.
The denser plot shows that there are essentially 6 branches coming off z=1 with angles pi/6 + k*2pi/3, k=0:5. However the computed poles for k=0 do not have matching angles (due to the multiplicity 6 and resulting large uncertainty on the computed value). This explains why the pairing is off.
