Heavy load peak when workflows run are often caused by large looping actions that consume all CPU from web front ends.

First thing to check is if there is safe looping enabled in Nintex general settings in Central Admin. If not I suggest to enable it.

Basically it allows only one loop within a timer job cycle. Loops will take a lot more time to process but at least a workflow won't be able to kill your servers.

Otherwise you'd need to check if you find workflows containing large loops and see if they cause the issue.

Would that be an option?

Best regards

Enrico