The idea: do as little work as possible, and use as little information as possible, in order to make a rate limiting decision. In the common case this means we only need an ingress OP_MSG requestId and moreToCome in order to produce a correct response (excluding OP_COMPRESSED cases for now). Once we have those bytes, we can "drain" the rest of the response buffer without copies or allocations, showing a 14% performance improvement on the workload introduced in DEVPROD-33860