I'd probably collect a week or 2 worth of stats for Disk Latency, throughput and IOPS for each datastore before and after. I'd also build a test VM and run some IOMeter tests with a wide range of loads to get performance from a VM perspective before and after.
Keep in mind that cache can be tricky to benchmark because it is transient and really works differently with day-to-day loads vs simulated loads. Also, cache will probably only help with reads, not writes. (if it is anything like the cache from other brands, I haven't used EMC specifically).
If you are adding the cache to address a problem with a particular app, collect stats specifically for that app.