Admittedly, I can't find any documentation on this specific scenario, but I don't see why it wouldn't work.
There may be other views, but in a sparse mode environment the RP is defined by the "ip pim rp-address" command being configured on all the routers. In truth, the RP router doesn't know it is the RP until it receives packets for sources to register or receivers to join. Whether it goes to one IP or another on the router shouldn't make a difference.
So if you have two loopback interfaces on one router and the RPs defined with ACLs, registers and joins for one group should go to one loopback and the other group to the other.
Hope I'm not steering you in the wrong direction and understand wanting to see the actual documentation. It is a little unusual but, again I can't see why it wouldn't work.