This makes sense in real (non-virualised) server, because you can optimise a few things, i.e. to put pagefile on separate faster disk (i.e. SSD), maximise transfer-speed (by using the outer cylinders of hard-drive), minimise access-time ("middle" of disk), prevent swap/pagefile fragmentation, etc.
How much sense it makes with VM depends on your storage-configuration. For example one of my arrays is Raid10/SSD where I always create small vdisk for every VM, and put only pagefile/swap-partition (with fixed size) there. This vdisk is very fast, and (according to my own experience) it is much better than over-commiting memory, because with its own pagefile/swap VM "knows" it is using swap instead of RAM, and I can controll how much it uses it (i.e. with "swappines" in linux). If I over-commited memory and counted on ESXi-swapping, VMs would know nothing about it and would try to use all memory without any limit (i.e. for disk-cache).
Furthermore, two general advantages of having pagefile/swap on separate vdisk come to my mind:
- can be excluded from backups (at least from those off-line)
- it is easier to adjust swap/pagefile-size (even on-line) by extending or adding one more vdisk, without touching VM system-partition
But this all depends on your own configuration/usage/etc.