I read the study, and I'll point out its issues first and then go into the data later.
The journal is Journal of Clinical Psychopharmacology which has an impact factor of 3.153, which is decent for a specialized journal. The study was performed by a team in Iran (which can be a red flag as it is a country known to tie job advancement to publishing a lot of papers), but that isn't enough to rule out any of the data offhand.
It was a clinical trial with 27 people in the 900 mg gabapentin group (at last timepoint), and 20 each in the placebo group and the 1600 mg gabapentin group. In the graph of withdrawal score over time, I wish they did statistics on the individual points/included error bars to assess the spread of results.
However, the 1600 mg/day group did have a pretty large reduction in withdrawal symptoms compared to the 900 mg group and the placebo group.
Now we get into the crux of your question: this paper challanges the accepted limits of gabapentin absorbtion. The first thing that comes to my mind is that that number is an average, and increased doses will be more beneficial in some of the population (say people with greater amounts of the amino acid transporters that gabapentin hitchhikes past the intestine with). The absorbtion data also was generated in a non-opioid dependant population. There is a chance that the decreased gut motility that opioids cause, results in gabapentin hanging out in the gut longer, giving more time for the transporters to clear out and take up more gabapentin.
I like this theory because the last two timepoints of the 1600 mg/day gabapentin group seem to regress towards the 900 mg group. One explanation for this is that the gut is normalizing with abstinence, and the excess gabapentin is being less absorbed.
However this all could be nothing. Look at some more papers, as there will be outliers in the scientific literature. This can be caused by something as innocent as hidden variables that people did not account for, or a poor study design, to research misconduct. Gabapentin is pretty common, so at least there should be a pretty big body of work to derive a consensus from.