Hey Shankara,
We in fact did just finish up using mountebank for a round of performance testing. In our case we found that the limiting factor was more in the tool we were using to generate traffic than Mountebank itself. In our case the tool maxed out around 270 TPS, with Mountebank staying fairly responsive throughout.
I will echo what Brandon said regarding --nomock. Disabling the mocking feature is definitely important, otherwise you will start to see some instability over particularly long performance runs.
We also found it very useful to use the wait behavior to add in some artificial latency. In our case we were trying to have Mountebank come as close to mimicking a production environment as possible. We first gathered end to end latency numbers on what performance was like in production. We then looked specifically at the average latency of the services that Mountebank was going to mimick. After we had that, we fed those numbers into our stubs, and Mountebank was able to give us a pretty good approximation of how our API was going to perform in production.
It was really successful, and helped us build a lot of confidence about the performance of our API.
Please let me know if you have any more questions, I'd be happy to answer them.
Thanks,
James