Hi JSLee -- If you are going to use this microstrip line to connect to a 3-D arbitrary structure, then using a volume mesher like HFSS is appropriate, and you want to make sure you are getting reasonable results before proceeding.
However, if you are considering only planar strucutures (including metal thickness), then you should be using any planar solver. Planar solvers are much more efficient and much more accurate (both!) when applied to planar structures. I usually recommend Agilent Momentum for unshielded and Sonnet for shielded (I work for Sonnet). The shielded approach has the advantage of being the most accurate. You can get a free SonnetLite (shielded planar EM analysis,
www.sonnetsoftware.com) that will solve this problem easily. When you download it and install it, do Help->Tutorials first. Then you will be ready to do this problem in about 45 minutes. If you have problems, put a posting on the Sonnet Forum. Some very capable people keep a close watch there and can be very helpuful. We often recommend using both Sonnet and Momentum to analyze the same problem when you have a high cost of failure situation. Why take chances "trusting" just one tool?
If you must use a volume mesher, I agree that the power-current definition is most likely to give a good answer. However, if you are going to use the Zo in a circuit theory (nodal, schematic) analysis, then you should use the TEM equivalent Zo. The TEM equivalent Zo will give the circuit theory tool (TEM) transmission line the same current and voltage at the transmission line terminals as you have in the EM analysis. One chacteristic of TEM eqivalent Zo is that as you increase frequency, the Zo first goes down, then it starts going back up. This has been verified by experiment. I can email you several papers on both the theory and experiment if you like. TEM equivalent Zo is what Sonnet calculates. And you don't have to worry about which kind of port to use, there is only one kind, and it is exactly calibrated. I can send you papers on that too.
As for the low frequency behavior of Zo, it might actually be OK. Remember that S11 is determined by Zo for an infinitely long line. You can simulate an infinite line by terminintating any actuall length 56 Ohm line in a 56 Ohm resistor. (You need a complex resistor for a complex Zo.) You will indeed have a high SWR. What you are probably doing is terminating a short length of line (a few degrees, perhaps?) in a 50 Ohm resistor and indeed you are getting a low SWR.