Okay, I just quickly walked through the scene, noting changes. First, I switched off the BIG EMITTERS layer, as I will not be using those at all. Next, I switched off the RED LIGHTS layer, as I will handle those later. Looking just at the SPOT LIGHTS layer, we see that there are 12 circular faces on this layer, in a single group, which means they all share the output of the 199W @ 12.7 SPOT LIGHTS material. Following is a sequence showing the initial changes.
spotlights only @ 199W, camera EV14:
camera @ EV8:
emitter @ 1200W:
Note that we are seeing reflections from the surfaces where they were lacking with the 80 fake fill lights on BIG EMITTERS enabled. I chose 1200W just estimating that each light should be around 100W. Next, RED LIGHTS is switched back on. Given that the RED LIGHT EMITTERS material is set @ 201W, the relative power outputs seen in this image clearly suggest that more than one mesh is using this material:
Furthermore, hiding both the SPOT LIGHTS and RED LIGHTS layers, just to ensure that there is nothing else using this material, we get no error from the render engine, indicating that there are still emitters assigned somewhere. Here they are, hiding inside the ceiling:
These faces appear to be on the spotlight fixtures. Removing the RED LIGHT EMITTERS material from 10 faces (3 each on the two end fixtures, 2 each on the middle ones), we obtain this error message from the engine when attempting to render:
1. There are no light sources in the scene (emitters, sky, IBL). The render will be black.
Which is exactly what we were looking for, meaning that the only objects now using the RED LIGHT EMITTERS material are the five groups on the RED LIGHTS layer.
This actually provides an excellent partial example of what it means when Maxwell calls itself unbiased. A naive approach would be for the engine to analyze the scene, guess that the camera can't see inside the ceiling, and just discard these extraneous emitter faces. The reason it would be naive is this: it is not actually possible to determine, analytically, whether you can see inside the ceiling or not. On the one hand, the volume may be a closed mesh, where given a material, we should be able to decide, based on the transparency of the material, whether the volume describes a "dead" space. However, this could only be proved in the most trivial of cases, and strictly speaking, not even in those (all textures and parameters would need to be analyzed, with arbitrary limits being chosen to determine whether various interdependent values are close enough to zero to be considered as being zero), given the generic nature of Maxwell's material system -- it does not describe materials in terms of their "transparency"; rather, transparency is something that may or may not result from the combination of components and properties you have used to define the material . And on the other, if the volume is not closed, it is always possible to imagine a more complex array of mirrors by which the camera is able to see inside the volume -- the only method of finding whether this is the case is the rendering process itself, and it does not yield a true/false result; perhaps you just need to render a little longer to finally find that one light path connecting the camera to a light source through a particular series of reflections. None of which is to say that you cannot in many cases make reasonable guesses about such things...but Maxwell doesn't. It does precisely what you tell it to do, and if that means rendering lights inside a ceiling, that is what it will do.
All that aside, it is unlikely that we want 1005W of red light (5 groups @ 201W each), as this is probably supposed to be some sort of strip neon/LED lighting. So, I explode those groups and join them into one, ballparking total emitter power down to 50W @ 26.5 efficacy (based on
this random reference, which quotes 26.5 lm/ft where you have about 50ft of lighting) to yield this result:
The reason for exploding/joining here is that the different surfaces differ in area, and would therefore require individual materials to be used, in order to obtain a uniform per-area result. Using Maxwell Studio or MXED to build an MXM, you could approach this by defining the emitter power in terms of lux (lumens per square meter) instead of total watts/efficacy.
I am not going to take it much further than that, just due to time constraints, but I hope you find this helpful. As noted, it's pretty clear, comparing this last image to the one you posted, that what was happening was that those 80 fill lights were simply killing all the highlights in the scene (and also necessitating a camera exposure in which the under-powered spotlights were not even contributing). I will leave it to you to create some other lighting in the scene, if that is appropriate for the setting.