I separate out Guest Operating System Pagefiles onto their own vmdk hosted on a separate VMFS datastore for the reason you mention. To save on the replication bandwidth. If the virtual machines are booted up at the disaster recovery site then a new guest operating system pagefile will be created so why waste the bandwidth replicating the data.
Depending on your backup solution, by separating the pagefiles out to different vmdks on different datastores may help you implement a solution where you do not backup the pagefiles.
I also do wonder sometimes if the added administration required to support virtual machines with pagefiles on non-replicated datastores if worth the saving we get.
Ideally you want your virtual machines sized for the workloads running within them so that they do not swap memory to the guest pagefile and therefore the pagefile stays empty and there is nothing to replicate or backup anyway.
I also configure the Hypervisor Virtual Machines Swapfile to be stored on a separate datastore, again so that I do not have to replicate them.