6.7 changed the counter that was used for CPU and Memory right sizing.
in versions prior to 6.7 i think it used the Active memory counter which is the active memory used on the ESXi host not what the guest was using so traditionally you could reclaim more memory and everyone's dashboards were green
Now 6.7 drops and it uses the guest free memory counter. This looks at the free memory in the guest as it is the safest memory to remove. This is one the 3 major counters for capacity & performance monitoring. The other 2 counters are Page-in Rate and Commit Ratio. In Windows, this excludes the cached memory. If this number drops to a low number, Windows is running out of Free RAM.
If windows runs out of free ram it will start to page which is where the Paged-in Rate metric comes into play. In windows if you click on the open resource monitor in task manager and select the memory tab you can see the standby memory and the free ( Application cache is the creater of standby memory). the Guest|Needed memory counter in vrops is the windows active + the standby/cache memory)
This change has been controversial at best but personally i like it as it gives me a better view of my memory landscape. It did however cause me to adjust some of the dashboards to turn them back to green as 90% of the VMs started showing a higher memory utilisation because it now took the application cache into account.
In your case i would have expected to see it the other way around (6.3 reclaiming more and 6.7 less) but just be aware you are not comparing the same metric