@Sukanya i see some problems with the above mentioned reasoning as -
Let us consider a scenario involving a typical layer 2 forwarding procedure where CRC predominantly does error detection .We have two major paradigms for forwarding at this layer - (i) store and forward (ii)cut through switching , store and forward is when the switches wait for full packet to arrive and cut through switching is when the decision is made quickly after the preamble and dest. MAC is read. So the transmit latency and buffering problem is irrelevant in store and forward architectures as anyways the whole packet is going to get stored . For cut through switching - ihe CRC is not even checked against and even erroneous packets get forwarded so cut through switching does not bring CRC into picture. If you look at the generator switch , it gets informaiton as a complete frame from network layer above ,in fact it gets a full packet which it converts to multiple frames - so clearly no such buffering and latency problems there also.