How granular should VM:s be for environments? [on hold]

How granular should VM:s be for environments? [on hold]

We are using Red Hat Enterprise Virtualization (RHEV) to create our
enterprise test and development architecture environments. We also have
several clients in the project, reusing most of the same code. I currently
have the following setup, where virtual machines are created on top of a
three physical machine cluster, and a physical machine is dedicated for
version control, CI, and VM manager:
Physical machine: version control, Jenkins CI server (master), databases,
RHEV-M
Virtual machine 1: Jenkins slave (client 1)
Virtual machine 2: Jenkins slave (client 2)
I will now create VM:s for several clients' test environments. Every
client has at least 4 test environments.
Should I create one VM for each client running many application servers?
Or should I create a dedicated VM for every application server? Does it
matter? Creating one VM for each AS would span at least 15 VMs, but at
least they would be easy to clone when creating new AS:s...