I highly recommend the aforementioned document to gain an overall better understanding of how memory management works. Hopefully, this blog post will help out as well:
http://www.vkernel.com/reader/items/vsphere-memory-usage-metrics-task-manager
Using the blog post as a reference, memory active and memory consumed at the host level are mostly an aggregate of the active/consumed memory of VMs on the host; therefore, memory consumed is frequently going to overstate what the VMs need from the host and memory active is going to understate what is needed.
Memory consumed is probably the safest of the two; however, you'll find that as you add more VMs, you'll miraculously find more memory as memory consumed gets de-duplicated via transparent page sharing (TPS) and idle/unused memory pages will get ballooned out of the virtual machines that no longer need them (VMs do not innately give back unused/idle memory to the host until there's a need and they're told to). When you begin to see consistent ballooning on a host, you you've most likely crossed a threshold for capacity where further memory consumption could degrade performance. In particular, there's a chance that consistent ballooning on a VM involves swapping inside the OS, so something to watch for.
Let us know if this helps or if you have further questions.