Trans per second on the report is Commands Per Second. That is collected from the Virtual Disks. That # is very accurate from a VM-perspective and I observe that repeatedly in many different environments from 6 VMs to 6000 VMs. If you're getting inaccurate readings for a datastore, I:
1. recommend ensuring you're actually collect proper #s from your environment - this includes hosts, VMs, and datastores. Check that you've got Read+StorageView:Views+Global:Health on everything, propagating from the root of vCenter.
2. If you think you're missing metrics, I can tell you point blank to check every single Host System in vCenter to ensure you're actually collecting performance metrics. 25-50% of the environments I see running 5.x Hosts have connection problems with at least 'some' of their hosts that are preventing metric collection - the Hosts connect and function, but if you try checking any performance stats you'll get 0-utilization or No-Data. If vCenter doesn't get the data, vCOps won't. People don't realize this until they start watching with vCOps and wonder why the data looks incomplete.
Let's step back from the Tps for a moment if we're talking about capacity.
The reason why you're saying that vCOps thinks one way or another is due to the way it has been configured. The vCOps capacity planning out of the box will always need adjusting for the environment in which it is running. This includes selecting a capacity planning model (demand/alloca), setting buffers and overcommit for cpu/mem/disk... all of section 3 in the config policy. vCOps CAN take in to account thin-prov disks and show accurate #s however you need to configure it properly.