I think you'll find they were all built with the same components, but electronic components have tolerances with which they are allowed to vary. The ASICs themselves will have some variation too, because at the sharp edge of processor implementation and design there are differences in the behaviour of individual processor dies cut from the same wafer.

Consider Intel for a moment. They fab a full wafer of Xeon chips intended for the high profit margin server sector. They slice the wafer and set each die in a substrate and wire it up, and drop the resulting chip in a test assembly. The test assembly runs each die at a variety of clock speeds and voltages, and the results of this testing identify a few dies as outright duds, and they are turfed. Some cores in a multicore die don't perform well, so Intel disables them and releases the chip as a hex core instead of an oct core. Some chips perform well on all cores, Depending on the clock speed and voltage each die will tolerate happily, the individual chips are "binned" as various frequencies. Some of the chips will run stably at crazy clock speeds, and these are binned as engineering samples or Extreme Edition CPUs, for which a huge profit margin is commanded.

The Bitmain ASIC chips will be the same. Some of them will go gangbusters and put up with ridiculous clock speeds without going wonky. Others will operate at more pedestrian speeds without failure. I seem to have a pair that will run at 500 MHz without complaint if I give them enough voltage and keep them properly cooled. I'm currently running them at 300 MHz because that produces the most work for my setup.

Philosophical waffling from me aside, if you have one 2Pac that has hardware errors at the same speed as another, and you have adequate cooling, turn the voltage up a little bit on the error-prone stick so it stops erroring. It's simpler than multiple cgminer instances if that's even a thing that works.