The vSphere client active memory measurement is using a sampling technique to estimate the amount of memory accessed in a recent time period.
It is an estimate based on a small sample size.
The OS is measuring the amount of memory used by the OS and the application that have been loaded into memory. Some of that may only be accessed in the initial boot - OS services and usually there will be other application bits resident in memory but have not been accessed recently.
So in most instances, "Active" memory as sampled by the hypervisor will be less than what the OS reports in use since the OS measurement is measuring a cumulative amount- or high water mark - at least that is how I think of it.. This is what allows memory saving and overcommittment. Both are useful, different time scales and different ways of measuring. Hope this helps...