To avoid feedback effects over the pressure (because it's a BC of pressure field). Try a simple test case with simulation 5, 10, 15 , 20 , 25 of the characteristic length. Plot CL versus characteristic length. In high speed simulations (transonic) we need even more than 20c.
There are generation of vortices due to fluid motion on lifting surfaces and the vortices are reponsible for pressure distribution on lifting surfaces. Pressure distribution on surfaces are responsible for lift and drag calculation. As we know vortices never dies, the effect of trailing vortices on the flying body with time diminishes. Conventionally, travelled distance 20 times chord length is an assumption, it can be more (agree with first answer). Refer to Low speed aerodynamic book by Katz and plotkin for better understanding. Thanks..
The key answer to your question is in the numerical scheme you use to solve the problem. Quite often the error propagates over the domain, therefore a larger domain is recommended in certain cases.