In few cases, VM's get SWAP Memory from DataStore (Instead of memory - from ESX DIMMs).
1) This could happen due to Multiple ESX hosts are down in cluster at same time (though abnormal),
2) Some time, Improper Sizing issues on Cluster level. (Design problems)
3) Unintended Memory reservations on Multiple VM's (May be by mistake / improper calculations)
4) Any other Memory issues on ESX hosts, etc..
Either of the above cases, ESX could not supply Real Memory (via ESX DIMM) to respective VM's.
Instead they supply Memory from "Datastore - Swap space"..
Previously we faced this issue, few VM's went into SWAP (memory from datastore space)..
But the real issue got resolved like "Recovered ESX hosts, Added Memory to Cluster, Adjust Abnormal Memory Reservations, etc".. By doing so, We should have enough free Memory on ESX Memory Pool / Cluster Memory Pool. But the VM's already taking memory from "Datastore SWAP", will not revert & they dont grab memory from "ESX DIMMs"..
In order to resolve this problem (VM's using SWAP memory from datastore), we had to Reboot the Specific vms, But now something is out there as per VMware (though it's not fully supported), but working decently.. Read at below URL for more details..