I have an extremely light go server that essentially either calls redis or calls an external rest service to generate a response. I think that because it’s so light, I’ll be able to set the “concurrent users” to higher than the 20.
Of course I want to test this… without taking anyone else down. Do I need to worry if I’m throwing 50-100 concurrent users at an instance with a shared CPU?
No need to worry! Go for it.
For reference, shared CPUs use Linux’s “completely fair schedule” to allocate CPU time: https://access.redhat.com/documentation/en-us/red_hat_enterprise_linux/6/html/resource_management_guide/sec-cpu
The micro-1x instances get 1/10th priority, and micro-2x get 1/5th priority under load. But if you’re running on a CPU without much demand (which is pretty common) you can burst up to a full CPU. Which really just means your benchmarks may look really nice. 
Interesting, thank you!