When merging a cube x,y,t,bands with another cube with x,y,bands, the result takes the intersection of spatial keys in both cubes rather than generating a union.
This piece of code is responsible for that:
https://github.com/Open-EO/openeo-geotrellis-extensions/blob/feature/spark3/openeo-geotrellis/src/main/scala/org/openeo/geotrellis/SpatialToSpacetimeJoinRdd.scala#L71
It originates from the fact that the code was intended for the mask process, where generating an intersection is correct.