You can't really say "I use FSX at the same settings", simply because you don't have ANY of the most demanding settings in FSX to begin with, like realtime shadows, cloud shadows, realtime reflections. Even when the slider looks the same, the sim is not doing nearly the same thing, like in Autogen, that P3D V4 draws (by default) using the much demanding Speedtrees.
I have much better fps in P3D V4 than FSX, even with Dynamic Lights enabled, and I have much better quality too.
The thing in the scenery that will affect dynamic lights the most, are the ground textures, assuming the scenery won't use its own dynamic lights, which only a handful use right now.
Since I don't like to make statements without actual testing, I made some tests, activated the display of the Average fps, which is the most accurate and useful to know so, these are average fps. And I tested with the default 737 imported from FSX, because I wanted to be sure I was testing the scenery, not the airplane itself. P3D V4 treats landing lights as a special case so, regardless if the airplane itself is "native" or from FSX, the landing lights are proper "real" dynamic lights so, I purposely used a not very demanding airplane, to put the airplane used out of the equation.
I don't think anyone would doubt that, if the problem was the scenery, the airplane used shouldn't matter, all other things being equal.
So, here's the results:
KCLT, which uses native P3D ground mesh/polygons, exported with the P3D SDK, sitting on RWY36C:
Landing lights On = 57.0 fps, with a 7.0% variance.
Landing lights Off = 59.1 fps, with a 5.0% variance.
KIAH, which use native FSX ground mesh/polygons, exported with the FSX SDK, sitting on RWY33R:
Landing lights On = 63.3 fps, with a 6.0% variance
Landing lights Off = 65.1 fps, with a 5.0% variance
LSZH, which use an FS8 ground mesh/polygons, exported with the FS8 SDK, sitting on RWY34:
Landing lights On = 54.7 fps, with a 5.0% variance
Landing lights Off = 54.7 fps, with a 4.5% variance
Note that, the most common case of airport sceneries, by far, are those using the FS8 ground polygons! We were always a bit different than the other developers, by using the native FSX format even for ground polygons, but you'll find that most sceneries out there still use the FS8 format.
But, as you can see from these tests, it doesn't seem the SDK used to make the scenery has ANY relationship with the impact of dynamic lights, surely not on our sceneries, but I cannot possibly guarantee it will be the same for every scenery out there. It's well known the FS8 code is slower, which is why we got rid of it years ago, starting with KDFW, but there's no evidence the FSX format is slower than the P3D one. KIAH it's just a less demanding scenery than KCLT, and that explains the fps difference between the two.
However, this test was made with Antialiasing at its lowest value, which is 2xMSAA. It's very well known that, more than everything else, what REALLY kills the fps with dynamic lights, it's the Antialiasing mode. So, putting that one at 8xSSAA, which is the better quality mode, the results become:
KCLT
Landing lights On = 29.1 fps, with a 4.5% variance.
Landing lights Off = 49.5 fps, with a 6.0% variance
KIAH
Landing lights On = 33.2 fps, with a 5.5% variance.
Landing lights Off =61.4 fps, with a 5.0% variance
LSZH
Landing lights On = 35.8 fps, with a 6.0% variance.
Landing lights Off =55.5 fps, with a 6.5% variance
But until now, I wasn't running with full screen, since I was writing this post at the same time. So, I went full screen, which in my case is 2560x1440, and see what happens, with the same highest quality antialiasing 8xSSAA:
KCLT
Landing lights On = 13.9 fps, with a 5.0% variance.
Landing lights Off = 31.1 fps, with a 4.0% variance
KIAH
Landing lights On = 34.8 fps, with a 4.5% variance.
Landing lights Off =14.9 fps, with a 3.5% variance
LSZH
Landing lights On = 17.2 fps, with a 6.0% variance.
Landing lights Off =29.8 fps, with a 6.5% variance
So, it's quite clear that, what kills performances with Dynamic Lights, are ANTIALIASING and RESOLUTION, which are basically the same thing, since increasing the antialiasing, it's just a way to force the video card to draw more pixels, as if it was running on a resolution higher than the screen.
The kind of SDK used to create the scenery doesn't seem to matter, not with our sceneries, which I chose to display 3 very different methods of creating ground polygons.