You could compensate for the speed differential by scaling out the slower annotator and using a larger pool in the aggregate so that multiple CASes are in play together. This doesn't change the latency of any one CAS but would increase throughput.
~Burn
