Hmm, how about doing what skinneelab guys do - several readings, dropping best and worst, averaging rest?

BTW, what different causes might cause different fan performance? If we theoretise/analise about it, maybe we can come up with better ideas how to counter those causes for more objective results?
Eg. how about ambient temp or humidity impact on fan rpm-s, storage circumstances (eg. storing time, vertical or horizontal) impact on bearing, bearing type specifics (eg. how do result repeatability looks for other ball bearing fans, do they fluctuate same as GTs?) and many other possible causes i forgot to imagine/think about ..