User rauno_s just gave a great suggestion on another thread regarding server testing. I figure I'd post my results of his suggestion to use Load Impact to test here. I've included charts of each test.

Virtual Specs:

1. 2 CPU - 4 GB
2. 4 CPU - 8 GB
3. 8 CPU - 16 GB

All tests were run using the same survey and having the application click through the survey. Each virtual testing server was hosted in New York via Digital Ocean SSD's using LimeSurvey 2.00+ Build 130802 and having MySQL hosted on the same server. VU’s were distributed equally from Ashburn, Chicago, Dallas, Palo Alto, and Portland. Each VU was using a generic “Load Impact Browser” to simulate this over an unlimited speed network.

Analysis:
CPU usage is the issue in this test, not memory. The 2 CPU’s survey handled about 100 Virtual Users (VU) before both CPU’s hit 100% and stayed there for the remainder of the 5 minute test. The 4 CPU survey test handled 125 VU’s before being strangled at 100% CPU usage for all 4 CPU’s. The 8 CPU run was obviously the best whereas the 8 CPU’s hit 100% 6 times and recovered very quickly (1-2 seconds) back to a 50-75% CPU usage range. 100% all CPU usage hit at 157, 175, 195, 215, 237, and 247 VU’s, but again they quickly recovered. User load time was predominantly under the 10 second window and hovered mostly at around 4 seconds for the 8 CPU virtual. The 2 CPU and 4 CPU virtual surveys were consistently above 10 second page load times once the CPU’s maxed out.

Charts:
2CPU - 4GB RAM


4CPU - 8GB RAM


8CPU - 16GB RAM

This thread has been bump by spammer, but it's still interesting.
Performance has been improved in LS3.
I'd be curious to have the same kind of benchmark.

The php/webserver setup was not mentioned in the thread.The impact of improvements of MySQL and PHP are huge. Are you aware of what load testing service was used in this test. Haven't seen this thread before. Thanks to the spammer.