Before we get down to the discussion, a few ground rules I hope everyone follow so that we can have an intelligent discussion.
Of course, no one is forced to follow them, but you won't get any points for being immature or trolling either.

1. Don't make baseless claims unless you really know what you are talking about.
For the purposes of this thread, this means you must have at least a little knowledge on modern graphics processing terminology, processes, issues, limitations and so forth to be able to discuss anything intelligently.
Also, this is more of a discussion for experienced modellers and I won't hesitate to go into the discussion of technical details.
2. Back up what you say. If you want to debate a point, take the trouble to do some research and post it here. This is easy as we live in the age of the Internet and Google, so there is no excuse for not giving proof.
3. "Lol I agree" is not an acceptable response; so discuss intelligently and people will respect you.
*******
Firstly, I will begin by saying that in general, model polycount is not the sole nor biggest factor that contributes to (or detracts from) the Almighty FPS.
Yes, I said it. Other factors do make a bigger impact on a game's performance.
To name a few (or alot...), dynamic per-pixel lighting and shadows, texture resolution, AI, pathing, maximum draw distances, physics systems, level of detail systems, graphic engine code optimisation/efficiency etcetera take a bigger hit on performance. Depending on the game and netcode, multiplayer games may also suffer due to graphics settings.
Other graphical bells and whistles such as High Dynamic Range (HDR) rendering, Light Bloom effects, Full Screen Anti-Aliasing (FSAA), soft shadows, depth-of-field effects, distortion/shimmer effects, dynamic reflections, particle effects, volumetric effects; these take the biggest toll on performance, and are usually optional toggles in games.
*However*, the caveat to all this is that most of the things I listed above are, in one way or another tied into the model's polycount. Upping the polycount will produce a noticeably higher load on performance with any of them enabled.
To prove I know what the hell I am talking about, proceed here and read through the discussion.
Standardised benchmarking on 3D cards/systems on credible review sources here, here, and here.
You may have noticed that all the games used in the benchmarking are FPS games; the reason being FPS games tend to put a greater load on the GPU and CPU, because many tricks used in RTS games to imply detail are simply not usable for FPS games, and pushes the hardware to their limits.
Here is another very good article on the subject of polycounts, for modding Alien VS Predator 2, an older but still a "AAA" title in its time.
Even though it deals primarily with level design, the fundamental concepts stated here are equally valid to this discussion.
I've quoted the most relevant parts for those too lazy to read the complete article (which has alot of irrelevant, AvP modding-specific details).
Source
Now, coming back to the issue of Spring.
Factors That Cause Levels to Bog Down
Levels can become bogged down for a number of reasons. The first of these is high polygon counts in a scene. Even a simple empty room that has only walls, floor and ceiling can be made up of several hundred polygons. [...] As a general guideline, the game designers for AVP2 kept multiplayer levels to around 700 polygons/scene in general on multiplayer levels and 800 or so for single-player levels where frame rate was not as vital. It├óÔé¼Ôäós important to perform some tests of your own on your target platform to develop your own standards.
[...]
The second cause of high counts is complexity. It├óÔé¼Ôäós very hard to resist the urge to put every detail into your geometry instead of using textures to create detail. However, it is much more expensive to add detail using geometry than it is to imply detail through tricks of the eye, lighting and texturing. Avoid nonfunctional detail that can be added through textures like molding/trim, curlicues, grooves and bumpy surfaces. A level designer's art is figuring out how to imply detail. Ask your texture artists to build a grooved surface or the face of a curlicued building ledge, and then apply that texture to a simple flat surface. Use as few sides as you can get away with on your columns and carefully consider how to build each shape with the fewest polygons.
"High" polycount models seem to have no effect in Spring, because reasonably modern cards have gone far beyond the industry-standard 1-2k poly models for RTSes just a year or two ago; current "AAA" titles such as Joint Task Force, Company of Heroes, Dawn of War and other AAA-class RTS titles easily have 2-5k polies (or higher) for individual models, and a ballpark estimate of 50-150k total polies onscreen at once.
Also note that these are also the last RTS titles to be released prior to the much-hyped Next-Generation breed of games.
Even an early model P4 with the equivalent 3D card can handle 1-2k models without too much trouble.
You may argue that many Spring players probably have decent systems already; as nowadays SLI and dual-core rigs are becoming commonplace.
But the point of learning to make the most of your polies, as opposed to throwing them around indiscriminately, is simply quality.
Modelling is an art, texturing even more so and it takes skill and finesse to do both well, and this in turn is achieved with practice and making every poly count (pardon the pun).
Poor, unoptimised models result in a messy lump that is difficult to UV map and a nightmare to texture.
That being said I can't honestly claim I am the best in the world at both, but as I have worked on the lowest of the low in terms of polycount, and relatively high polycount models throughout my time as a modeller for both RTS and FPS mods, besides reading up on the ins-and-outs of game modelling, I believe I can consider my essay thus far is reasonably authoritative on the matter.
And if you're still alive and awake after reading the entirely of this essay, then congratulations, thank you for bearing with me and I hope you have learned something from it.
