Why is ScaleNU 20 times slower than Scale?

Just a general question to understand and hopefully work around it in some way. If you have larger datasets you notice some components taking unusually long and its always good to find out why to find an alternative way of solving a problem.

Here I have a lot of compound transforms (1 translation and 1 scaling) and apply them to a simple cube. In theory (and from other software) you would assume that since it is compounded into a single transformation matrix that it should not take any longer or shorter, but for some reason Scale NU (non-uniform) takes at least 20 times as long as a uniform scale.

Can anyone explain why that is happening and if there is a better way?

scale_nu_slow.gh (155.3 KB)