Optimizing/Capacity Issues - Composite Models

I have a question about composite models in regards to optimization and capacity usage. Some of our data analysts have recently discovered that they can link to an existing dataset and then create a composite model by clicking the option to "create a local model" and add in some data from somewhere else (for example, some Salesforce tables or some Excel files). We haven't been having any issues with slowness, reports not loading, or refreshes failing for over a year, so I'm suspicious that the recent increase this practice might be a bad idea. Specifically, yesterday there was a sudden spike to 500% of our capacity that was all down to one specific such model (I was able to track it down both via the Capacity Metrics app and by talking to the person working on that dataset/report). In this case, the composite model is from our largest and most active dataset and several tables from Salesforce.

What is your experience with such composite models? Is it likely that this is the reason for it causing such a spike, or is it more likely to be something else that is wrong with the dataset/report? There are no measures at all defined within the dataset that aren't native to the original dataset.