You should try to combine them into the same mesh as much as you can. As long as they are in the same Geom object, they will be in the same buffer, and sent in the same batch to the GPU. Your performance will otherwise start degrading pretty rapidly once you hit a few thousand Geoms.
Panda provides a flattenStrong() method that you can call to try and combine as many geoms under a node together as possible; but it’s still better to have them already combined in the first place.