Memory is pretty complicated, to the point where recalling memories makes them labile, destroying connections is as important as making connections, there are different processes underlying different types of memory, etc. So it's possible all of these results are replicable in various contexts.

One thing that helps with biology papers is mentally substituting "performance in our mouse model of memory" for memory, they are using some very specific metric for memory in aged mice. It may be that this result doesn't hold true for a different memory metric, in different animals etc.

Also, without reading the full paper it's unclear whether THC harms this type of memory task memory in young mice, which could totally be true.