I came across a paper hep-ph/9411381 suggesting the use of three SO(10)'s hitting distinct 27's of E6, in order to properly respect triality. The use of three 27's seems to take one back to the maximal subgroup E6xSU(3)/(Z/3Z) of E8, giving the decomposition: 248=(8,1)+(1,78)+(3,27)+(3*,27*)...