Hi Samuel,
Clusto is actually designed to work the other way around: you use clusto as a source of truth and make the rest of your infra mirror whatever "truth" is inside clusto. For example, instead of having clusto query mesos for a list of slaves or jobs or etc, you model your mesos task in clusto (by, for example, writing the drivers that can model that sort of thing) and then mesos would adjust appropriately.
Given the use case you mention: I would model my VMs in clusto with a driver (I haven't heard of a clusto-vmware specific driver, so most likely you'd have to write it yourself) and then control them via .create() or .destroy() methods inside it. Does it make sense to you?
Regards