The polymorphism is found at its highest frequency in Southern Europe and Southwest Asia, but Razib argues in favor of its ultimately African origin because of its existence at a low frequency in the "isolated" Mbuti:
If H2 arose in the Middle East its presence in Africa could be explained by back-migration. I immediately was skeptical of this model because H2 is extant at frequencies of 5% among the Mbuti Pygmies. The Mbuti are relatively isolated genetically from the Bantu farmers who have come to dominate their region. If there was any group which represented the ancient genetic variation of Central Africa, it is likely the Mbuti.While the Mbuti are indeed fairly isolated, it is not right to think of them as a relic of African genetic variation. The distinction is crucial: while the Mbuti contain a large component of "ancient" (pre-farming) African genetic diversity, they are by no means pure, as can be seen by last year's comprehensive study on African variation.
In other words: the Mbuti are a good population to find ancient African variation in; but, they also have admixture with more recent populations, which may have served as a conduit for the introduction of extraneous genetic elements such as the inversion in question.
Getting back to the paper itself, the moral is clear: be doubtful of genetic age estimates for extremely old events (by which I mean anything predating the Neolithic).
This includes the age estimate in this study which is based on microsatellites. The long-term behavior of these markers is based pretty much on conjecture, as we normally have the ability to observe them only across a few (human) generations.
So, I'm not taking as granted the accuracy of the current paper's age estimate. Nonetheless, the study of the distribution of the inversion is quite useful. Its origin is probably in West Asia, a region which would explain its presence in both Europe and Africa.
The Distribution and Most Recent Common Ancestor of the 17q21 Inversion in Humans
Michael P. Donnelly et al.
The polymorphic inversion on 17q21, sometimes called the microtubular associated protein tau (MAPT) inversion, is an ∼900 kb inversion found primarily in Europeans and Southwest Asians. We have identified 21 SNPs that act as markers of the inverted, i.e., H2, haplotype. The inversion is found at the highest frequencies in Southwest Asia and Southern Europe (frequencies of ∼30%); elsewhere in Europe, frequencies vary from less than 5%, in Finns, to 28%, in Orcadians. The H2 inversion haplotype also occurs at low frequencies in Africa, Central Asia, East Asia, and the Americas, though the East Asian and Amerindian alleles may be due to recent gene flow from Europe. Molecular evolution analyses indicate that the H2 haplotype originally arose in Africa or Southwest Asia. Though the H2 inversion has many fixed differences across the ∼900 kb, short tandem repeat polymorphism data indicate a very recent date for the most recent common ancestor, with dates ranging from 13,600 to 108,400 years, depending on assumptions and estimation methods. This estimate range is much more recent than the 3 million year age estimated by Stefansson et al. in 2005.