Is there a difference in computational efficiency/performance between loading a heightmap in Gazebo vs. loading an equivalent static mesh as the visual and collision elements of a model?

It seems that meshes generated when gazebo loads a heightmap are cached, which I don't think is the case for standard geometry(?), so that would at least seem to indicate that heightmaps might be preferable when sticking with a single terrain model. But are there any other differences in how they are used under the hood?

In particular, I'm interested:

• once spawned in gazebo, heightmaps appear to be converted into meshes; are there any difference between how the meshes are used?
• if heightmaps perform better for larger/higher resolution terrains during the simulation? while loading the simulation environment?
• and if either would be preferable when simulating a large number of randomly generated terrains (e.g. where each run of a simulation has a different terrain loaded)?

Thanks!