What is the main difference you need to use a scene capture cube for? Have you tried instead increasing the FOV of a scene capture 2d? If you aren’t using it for a full stereo reprojection it might work. Also you can turn down the resolution of the cube render target but I am pretty sure the main perf issue is that scene capture cube will actually use 6 scene capture 2d operations. So this means your CPU tasks will be done 6 times including things like occlusion and culling.
Have you tried using the “Show only actors” filter to list only the bare minimum scene actors for the cube? Doing that and keeping the count very low is about the best I can think of to improve the speed without doing some code changes of some kind. Basically it should lessen the CPU burden to tell the engine which meshes to render instead of it having to figure it out 6 times, but I haven’t tested this for performance yet.