There are technical and operational considerations.
The technical consideration is anticipating or determining how much load the RR platform will be subjected to. Are you going to be carrying 100>1000>??? of routes? Will there be lots of churn? Generally, RRs obtain a complete copy of the BGP RIB. So if you have a churn event, it propagates to all of your route reflectors. If they are also in the forwarding path, you can wind up with problems in your data path. Specific to the MX80, the RE isn't the best for RR services with lots of routes. I wouldn't combine them.
Operationally, are you OK with fate sharing your RR and your data plane? Consider: maintenance, change policy. If you don't have a change policy, are you hiring? Kidding.
I can tell you that in the past I have combined my P node and RR. I did so on a larger platform but I also lsys'd the box so that the RR portion of the router could lift right out when I decided to go to discrete RRs.
If I had a greenfield to build now, I would either get stub 240s with dual x64 REs or run it on a server running something like Vyatta. In some topologies, I'm not even sure I would put the RR in the IGP. One less thing to go wrong.