Anecdotal observations like these are tantalizingly interesting.
Perusing some recent benchrest match group sizes at 100 and 200 yards shows that about 80% of the groups grow in average size (in MOA) between 100 and 200 yards, and about 20% of the groups shrink in average size. Only 10% of the groups shrink in size by an amount that can be considered statistically significant.
When only 10% of the available data shows the shrinking group sizes with longer range, the question becomes whether this observation can be dismissed as a result of random chance, or whether it is systematically produced by some feature of the rifle barrel, bullet, and conditions. Anecdotally, we have noticed that the feature of shrinking group size seems to occur commonly for combinations that are only weakly stabilized, such as a 53-55 grain .224 bullet in a 1 in 14" twist.
Unfortunately, the available benchrest data on group sizes does not contain enough detail to analyze the data with some meaningful selection criteria that would allow assignment of a possible causal factor to the shrinking groups at longer range. To compute stabilities, we would need a record of bullet make, model, and weight, barrel twist rate, muzzle velocity, ambient temperature, humidity, and atmospheric pressure.
Consequently, we cannot yet determine whether or not the occasional occurance of smaller group sizes at longer range is a meaningful and predictable consequence of definite factors or whether it is the result of random chance.