No, I did not test DX-Studio. I have read through the feature list and did not find any technique like occlusion culling, portals or similar to achieve something bigger than a small scene. So I just went away.

The results of playing around with my corridors and textures in different engines are:

C4 rendered fast and looked very good. Zones and portals work perfectly but there can be slight events of stuttering when you turn around close to portals. Because of the forward renderer you need to setup lighting with lights using small radius and no or only little overlapping. Because of this and because of the collision mechanics it is better to split meshes into several parts. Big meshes will be influenced by more lights causing more draw calls. The interior shadows are stencils, so you need closed meshes and you get sharp shadow edges.

Unity provides both: forward and deferred rendering. The deferred renderer allows me to use as much lights as I want. So I can create more realistic lighting. But the deferred rendering creates way too much draw calls. The same scene with forward rendering uses only a quarter of draw calls. The same applied to the displayed amount of polygons. I had only a couple of thousand polygons in view, but the deferred renderer produced more than a million polies. The reason is probably that the scene gets rendered a couple of times (once for each shadow casting light, once for the depth map and so on).
But the occlusion culling works perfectly in Unity. So it also renders only what you see.
Movement with the included character controller produces stuttering even when I have enough FPS.
For collision detection I needed to add extra low poly geometry otherwise the CPU gets heavy work and there is no multi-threading.

I had better results in UDK. It renders amazingly fast and movement is very smooth. The shadows are baked but the normal and spec mapping works perfectly on top of the static lighting. The latest updates also added a lot of real-time lighting effects.

But in the end I was suprised about the power of the Vision3d engine from Trinigy. It can do the same like UDK, Unity or C4. It can render in forward and deferred mode and it can use directional lightmapping using static shadows the right way with shaders. It can use shadows via stencils and shadow mapping via shaders.
It calculates static lighting information into cells to be applied to moving geometry if needed. The lightmap compiler is the fastest I ever saw even with multisampling and radiosity activated. And the renderer is very optimized. It splits geometry automatically into mesh groups using the same materials to reduce draw calls. The editor is way more intuitive compared to UDK or C4. It can be compared best to Modo with its different layouts and panels.

Torque3d failed because the zones and portals are not working and I experienced other bugs.

So in the end it depends on the project again: I would use Unity for everything small just because of the easy scripting and the number of available platforms.

But if I need the power for bigger game maps I would probably switch to Vision3d (if I had enough budget). UDK and C4 are more cheaper alternatives.

By the way: UDK also supports iOS and Android now and Vision3d just got an iOS port and a web-player (besides all the available consoles ports). So there are lots of options today to go multiplatform.