The latest revision of the path antennas arrived.  The network analyzer indicates that the patch resonance frequency is no correctly centered at 1.420 GHz, but the impedance matching is still incorrect.  Working backwards from an input impedance of 86-40j Ohms with a 147.6 Ohm quarter-wave TL yields an impedance of 208+97j at the patch edge.  Changing the TL impedance to 112.8 Ohms (20 mils) reduces the input impedance to 50.2-23.4j.  The imaginary part can then be canceled out with a 2.7nH series inductor.  I've submitted the revised board to OSH Park.