I'm afraid this is not just an outlier, but a rather typical situation. We don't know anything about how often people played on the various venues. In all likelihood, there will be many entries that are only based on very few games, or where a played started on one venue, and then continued on the other, while improving, and the like. We simply don't have enough data, which naturally causes data points to spread out. KGS and IGS happen to be the ones with most data, so the fact that this graph looks largely exact is probably the exception, not the rule.

BTW, I forgot to include Nick Jhirad's value: 5d @KGS, 5d-6d @IGS.