Here's some empirical data from one of the welds that I've had hooked up to data acquisition systems. I've got a weld that's mechanized pulsed GTAW, about 250 amps over 175 amps (primary and background, respectfully). Voltage is about 10.5v for the 250 amp pulse and 9.3v for the 175 amp pulse. Same arc length.
Now, if you hooked this up to an arc voltage controller (AVC) that monitored voltage all the time, the torch would move with every pulse. Lets say I set it for 10.5v. It'd be in one location for primary pulse, then it'd move the torch to make the arc longer during background pulse (longer arc, more voltage dropped in the arc). This is generally not desirable in mechanized GTAW, but I've seen procedures with it. For this reason, some equipment (I use alot of AMI stuff, so I'll use them as an example) will have different options. Sample primary (read's the primary voltage, ignores the background voltage). Sample background (reads the background voltage, ignores the primary voltage). And AVC Continuous which reads voltage all the time. Sample primary and sample background allow you to just read voltage during that part of the pulse which when setup properly, more or less keeps the arc length constant from pulse to pulse (barring other odd stuff that can mess with arc length).
The weld data above was performed with sample primary, so the arc length did not change between primary and background pulse.