I imagine if you’re dealing with 2k on the smaller axis it won’t matter too much. however you’re probably “losing” information encoding it down to a square, or wasting memory/bandwidth, depending on if you err on the small size or larger size.
I think equirectangulars are 2:1 because the image on u wraps around the spheres entire circumference, while the image on v goes only from north pole to south pole of the environment sphere.
Also, environment lights rely on internal mipmaps to generate reflection maps for different roughness levels of a material . So, whatever res you go with ideally you’re using mip map friendly sizes like 1024 , 2048, 4096, etc. Though you can certainly get away with odd resolutions, the lowest mip levels can get weird.
Another way to look at it is a cubemap also represents a “sphere”( imagine a poly smoothed cube, the 6 faces would be spherical) and so I think the top that is converting the cube to equi would be ideally set 2:1 to capture consistent detail density on both axis