Creating a new BinReader for every instruction is a bit too much and it adds about 1% overhead on block import (and actually is quite visible in the VM profiling statistics). So use a bit more ugly but efficient method.