You seem to be speculating that an exchange of ions between the analytes and whatever's in the mobile phase might be frozen out if there's < 5% or so of water. That's a falsifiable hypothesis. If you're game to indulge in a little Truth-and-Beauty, then here's an experiment you could perform to test that hypothesis. Take a charged analyte that has one counterion (because that's the counterion in the sample solvent) and inject it into a HILIC column where the mobile phase contains a very different counterion of that charge. If there is counterion exchange, then your analyte peak will either be severely skewed or will elute in two separate peaks with a continuum between them. One of those peaks will be the ion pair consisting of the analyte with its initial counterion. The other will be the ion pair of the analyte with the counterion supplied by the mobile phase. Since the two counterions differ in polarity (since you chose them that way), the ion pairs elute at different times in HILIC. The continuum will be molecules of the analyte that started with the original counterion but which exchanged it for the counterion in the mobile phase during their migration down the column. There's an example of this pattern in Fig. 14 of my 2008 paper that introduced ERLIC, cf. the following link:
http://pubs.acs.org/doi/pdf/10.1021/ac070997p . Here, the analyte was arginine, with an anionic counterion that either did or did not match what was in the mobile phase.
If no counterion exchange occurs at high % organic solvent, per your speculation, then you won't see this two-peak pattern.