Why do the computer programs use both 100% and 85% arcing current in determining the incident energy?

There was a minor revision / enhancement to IEEE 1584 in 2004 that added the 85% At one time the calculations required the input of "bolted" short circuit current data i.e. no arcing impedance that would reduce the amount of current that flows. Then IEEE 1584 developed a method for estimating / calculating the arcing short circuit current based on a number of tests. The significance of the arcing current is so you can look at a time current curve and determine how long it takes an overcurrent device to operate based on the current during the arc flash. After the 2002 1584 was issued, there was concern that "what if" the arcing current was a little lower than estimated. That could make the difference between whether the device operated instantaneously or timed out in the overload region - having a huge effect on the incident energy. Now you evaluate it at 100% and also 85% to see if 85% causes a longer clearing time that would lead to much worse incident energy. The worst case of 100% vs. 85% is used for the final results.

"there was concern that "what if" the arcing current was a little lower than estimated"

What are the reasons for arcing current to be lower than estimated?

Many factors can change arcing current... lower available ultility fault currents, voltage fluxuations, cable impedance and cable run distance, etc.

By its definition, an arcing fault is not a bolted fault, and the arc has resistance as well.

My interpretation of the 1584 amendment is that the 85% arcing current calculation is only intended for use on <1kV and that when performing calculations >1kV, 100% should be used.

That is the general concensus for the interpretation. Above 1 kV values of Ibolted and Iarcing begin to converge with only a small difference between the two. I believe most (all?) software also use this interpretation.

In case it may help someone else, I'll note that at least one software is having some trouble with applying the 85% correctly based on voltage.

The SynerGEE software v3.8.3.153 by GL (formerly Advantica) calculates arcing current at both 85% and 100%, but the summary report only displays the resulting incident energy that is greatest, so it could be from the 85% or 100% value, which is okay for <1kV, but not expected for >1kV. There is a separate report to see the details, the problem is with the summary report.

Another problem is that the software appears to have an option to modify the reduced arcing current from 85% to any specified value, but this setting does not actually do anything at this time. You would think that you could force the calcs to 100% using this setting when studying >1kV, but it currently does nothing.

By definition an accident is undefined; one does not know the length of the arc, its plasma content; therefore what the true arc current will be is also unknown. There was a paper given at NFPA in Boston in which an engineer discussed the problem of a generator when the Arcing current was low enough that it resulted in a long time to clear resulting in very high heat releases.