Efficient techniques in L2-optimal model reduction