That depends on what the app is doing, and how it's doing it. 50% is the theoretical maxium for 2 machines instead of 1; 0% would be the minimum. Amdahl's Law is a rough guideline to what's achievable.
Unless you are talking about load balancing a process to multiple machines (efficient use of cores), I don't see how performance will get impacted by load balancing.
If you are load balancing traffic coming into web application, it improves availability and not performance (This statement assumes that you have enough resources to handle set amount of traffic on all machines).