Sounds like @bghgary has solved this but just wanted to add that we encountered some performance challenges implementing multi-canvas views also.
It turns out using different sized canvases forces a resize on each view render so it’s important to use the same size canvases.
Also, @Evgeni_Popov kindly added the ability to disable a view for us, so we can use intersection observers to enable a multi-canvas view only when it’s visible.
We also use intersection observers to only load the 3D after it scrolls into view, but that was mostly for SEO/page rank/speed reasons.
See …