There are three answers: on the primary machine, on the replicate machine or on a completely separate machine. There is no right answer, and if you are doing an initial install it probably pays to consider the future, consider the proposed configuration and have a look at the load on the available machines.

It is probably fair to say that replication is not power hungry but neither is it free. If the primary is only just about coping with its current load, then it might be as well looking into hosting it on another machine. The argument applies to the replicate. If you think that network bandwidth may be an issue, and you may have to add a second replication server, you may be better off starting with repserver running on the primary. It is marginally easier to add a repserver to an existing configuration if the first repserver is on the primary.

Remember that a production replication server on Unix will require raw devices for the stable devices and that these can be more than 2GB in size. If you are restricted in the number of raw partitions you have available on a particular machine, then this may have a bearing.

Installing replication server on its own machine will, of course, introduce all sorts of problems of its own, as well as answering some. The load on the primary or the replicate is reduced considerably, but you are definitely going to add some load to the network. Remember that ASE->Rep and Rep->ASE is uncompressed. It is only Rep->Rep that is compressed.