Plugin lifecycle: update() vs renderer() best practices

[JonDum]

I'm heavily invested in making custom plugins for lwc but it's been nagging me what the best most performant way of writing them is. There seems to be many discrepancies in the plugin-examples folder and I'm struggling to see what the whole point/benefit of having an entirely separate renderer class is when it seems like any trivial action like scrolling or scaling the chart will call the plugin updateAllViews() which in turn calls each paneView's update() right before renderer().draw() is called.

It seems like all the calculations like timeToCoordinate() that are needed for the renderer are supposed to be calculated and cached in the paneView and passed to the renderer, but most of the time all these calculations are changed for every render anyways, so where's the benefit? Every single render()#draw() seems to be proceeded by update() regardless.

Even worse, many of the plugin examples seem to return a whole new class instance in renderer() (example)

And this gets called constantly, even as the mouse is moving around. At the very least I would have expected the renderer to be cached and reused instead of constantly getting recreated, no? In some of my plugins I instead create the renderer once in the paneView constructor and return the cached version. Seems to work fine, and that is certainly going to be more easy on the garbage collector than making thousands of new classes on every tick. So what am I missing here? What was the thinking behind this plugin architecture and what are the best practices?