Some of the Ants here had loose screws holding the printed circuit boards to the heatsinks. Others had excessive amounts of thermal paste (and loose screws). Get the right size Allen key, and gently check all those screws (in particular, the one at the center of the boards). If you find any that are loose, tighten it to 'finger tightness', don't over-do it. Tightening any loose screws will also 'squeeze' out some of the excessive thermal paste, which can be simply wiped away with some tissue paper. There is a single temperature sensor per blade, in the center, so if one of the printed-circuit boards that is far from the sensor was not attached properly (loose screw(s) and way too much thermal paste), the ASICS that suffer from over-heating due to poor thermal conduction may go into thermal shutdown or simply display excessive errors due to this 'local' high temperature which is not seen by the sensor that is too far from that area.
If this doesn't help: Try swapping the power supply with one from the 'good' Ants - maybe there is marginal performance or current capacity. The next thing to try is a different clock speed. My impression is that every Ant is different, some 'like' higher clocks and some 'prefer' lower settings. Try changing this in small steps, with at least 10 minutes before things settle down and you get a representative reading. You should consider the hash rates, HW error % and power consumption, together with the X's. Look at the overclocking thread
https://bitcointalk.org/index.php?topic=348327.40 or other guides for instructions how to change the S1's frequency settings.
Good luck
