running a separate database (mongo) for a subsystem instead of using the current one (cassandra)

I have the following scenario. There is Cassandra in use and is expected (in the future) to get a lot of data in it. It is supposed to scale totally horizontally. I have a second subsystem that deals with user messaging service. This can exist in mongo for example in order to separate the subsystems. But this means using up more cpu and mem cycles for data that could be stored in Cassandra.

Is it better to keep less data in cassandra and have the overhead of running two databases or just stuff everything in it.