It all depends on whether the SQL server data volumes will be VMware virtual disks or Raw Device Mappings presented to the VM.

Virtual disks act as separate entities and can be pinned to separate ESX data stores, the problem is that these data stores usually service virtual disks from lots of other virtual machines of possibly differing types (exchange server, file server, application server, etc). Virtual disks are flat files and do have their limitations especially when hit with random I\O workloads.

Be mindful of presenting too many virtual disks, as these disk require a virtual disk controller, this itself is a software representation of a physical disk controller and runs as a software process on the host server. All of this takes valuable CPU time and RAM resources from the host.

Raw device mappings are straight to the storage layer and are managed by the storage layer itself, not the software disk controller. They'll generally handle random I\O a little better but even so, with any virtualisation exercise you need to account for the fact there will be an overhead over the physical equivalent.

Remember, virtualisation is a consolidation technique not a performance technique, when done properly you can attain sensible performance levels. Get it wrong and you'll have even more smoke to cloud your vision when attempting to troubleshoot