[PERFORM] Impact of track_activity_query_size on high traffic OLTP system

I have found some examples of people tweaking this parameter track_activity_query_size to various setting such as 4000, 10000, 15000, but little discussion as to performance impact on memory usage.  What I don't have a good sense of is how significant this would be for a high traffic system with rapid connection creation/destruction, say 1000s per second.  In such a case, would there be a reason to hesitate raising it to 10000 from 1024?  Is 10k memory insignificant?  Any direction here is much appreciated, including a good way to benchmark this kind of thing.