What is the limit of CFD by the Simulation of the flow through a microchannel?

(1). There is no limit on CFD simulation. (2). The possible limit is likely related to the hardware (computer). (3). Whether the cfd results will match the test data, is a different issue. Even the test results are not always accurate. The accuracy issue is related to the correct modelling, solution algorithm, and the mesh used. If you are not using the right physical models, the correct solution algorithm, and the systematic mesh refinement, then you will sure have the accuracy problem. But these are not the limit of CFD. The same problem will happen when the testing is not done properly. (4). But if you need the fine mesh and the hardware can't provide the needed memory, then this is one of the limit on CFD, a hardware related limit.

Ofcourse there is a limit. Most (if not all) CFD codes asume some kind of continuum phase. If you go to very small scales then you approache the Knudsen flow (or free-molecular flow) regime. Here the average pathlength of a molecule between two collisions is of the same order as the geometry (or particle). This is the area of rarified gas dynamics or monte carlo simulations where you have to model the particles (molecules) individually. This is at least one (practical) limit. But in general: If you find a model that describes it, you can solve the equations and get a solution.

But this cannot be very helpfull, I think. Every code has its own limitations.