Hi:
When you use a lumped port, I believe that the 3D simulators are usually creating an ideal "wire" from a location on your transmission line (or whatever body you are attaching your lumped port to), to whatever you choose to define as the ground for your port. Somewhere on that wire, they introduce an ideal source, which might also have an internal impedance that is defined.
This creates a discontinuity on your transmission line because the current (which at higher frequencies flows primarily out on the edges of the transmission line) must now flow around the open end of your transmission line in order to funnel into the "wire" that will be used for your lumped port. This creates local field and current disturbances, which will lead to reflections of signals back down your transmission line.
If you use a waveport, your termination looks different. It doesn't have the same discontinuity. Good wave ports will terminate your exact field distribution more accurately (I'm assuming you only have one mode of propagation, even though some tools like CST have the ability to make waveports that terminate as many modes as you like) and will not have the added disconuity that arises from having to disturb the natural current flow in order to make the connection.
I would think this could be easy to test with a length of stripline transmission line; make it at least a quarter-wavelength long so your ports don't cross-talk to one another. You can find papers that describe the exact theoretical impedance of a flat symmetric stripline. Make one test through line 2-port problem with a waveport and another with lumped ports. Analyze the structure under perfect termination conditions (that is, make sure that your lumped port uses an internal impedance that matches the exact theoretical impedance of your transmission line).
You will probably see some phase differences between the two, but I would especially check the |S11| and |S22| data. I am guessing that the data will be much lower (i.e., less noise and less reflected power due to port discontinuities) with the waveport than with the lumped port. This will also give you some idea of how much error your lumped ports add to your simulation.
Also, if you sized your line to be exactly 1/4 wave at, say 15 GHz, I'll bet you will find that the waveport version will show close to 90 degrees of through phase (ang{S21}), where the lumped port might show you a bit longer phase. Try this one out--it would be interesting to hear from someone who runs this.
To answer the question of manual de-embedding lumped ports... I think it would involve running three different EM analyses for each unique lumped port. Think of this problem like you would if you were making a real microwave measurement with your VNA. You would have to measure your calibration standards, right? So you might perform a TRL cal, or some other variant (there are several). You can probably find a good application note on making calibration standards and de-embedding theory on nearly any VNA manufacturer web site (Agilent, or R&S or Anritsu or others). For each lumped port, you would repeat the exersize of creating EM simulations of calibration standards in order to come up with S-parameters (or ABCD matrices) to represent the lumped port discontinuity, and then cascade the inverse of these parameters with your original structure simulation using netlist (or schematic) simulation.
Here is another "poor man's" way to manually de-embed the ports:
Create a 2-port problem for your transmission line (with the lumped port attached), and simulate it. Take your S-parameter data set to a linear circuit simulator. Create a model in your linear circuit simulator of an ideal transmission line that matches the line you used in your EM simulation, and to one port attach a series L, and a shunt R and C pair. Optimize the values of L, R and C until you get a model that yeilds data that is identical to the EM analysis results of the structure. The L, R and C are your "port discontinuity model." Once you have this model, then you can use it as a "negate" block to cascade with your lumped ports from your 3D EM simulations.
For the amount of work that this is, I would rather use a waveport where possible.
I hope this is helpful to someone out there.
--Max