I might be completely off, but these are things that seem dubious to me:
1. Is it necessary to multiply bone matrix with bone offset matrix? I would probably simply multiply with offset matrix assuming that before passing to glsl it is already multiplied with parent offset matrix (that is in turn multiplied with its parent offset matrix and so on).
2. Is that the sum of aVertexWeights 0 to 1 guaranteed to be 1.0? If not, maybe divide finalVertex with sum of aVertexWeights.