Tensored vs ordinary measurement mitigation runtime

My experiments show that the runtime of actually applying the measurement filter matrices to circuit execution results is much slower for tensored mitigation than ordinary mitigation. Tensored mitigation quickly becomes almost intractable as circuit gets larger.

I thought since tensored mitigation only accounts for local measurement errors, it would be faster to compute the matrices as well as applying them to circuit results? I am using measurement mitigation as described
here.