I'm just not expecting 40 fps difference between comangia's cfg and mastercom's comp preset. Here's the result from comanglia's cfg, as expected:

2639 frames 17.370 seconds 151.93 fps ( 6.58 ms/f) 14.589 fps variability

What guessing?
Switching configs isn't as seamless as I'd like it to be, I'll go with mastercomfig's comp preset for now, and I'll test comanglia's in free time, maybe later today :)

Yea, I was a bit in a hurry publishing these results, so here's more:
(forgot to mention: I'm also using the explosion mod, the no bullet dust mod and the no soundscapes mod - applies to the previous post as well)

compquality didn't really cause any gains:

2639 frames 20.215 seconds 130.55 fps ( 7.66 ms/f) 9.262 fps variability

comp, however, was significantly better:

2639 frames 16.819 seconds 156.91 fps ( 6.37 ms/f) 11.698 fps variability

Given the performance boost I got in CS:GO, I expected a bit more off of TF2, but oh well.

I'm thinking so too. In fact, I have a GeForce 210 I could also use to benchmark TF2 on, just for science...

Another occasion for another experiment. This time I made use of the ongoing GPU prize craze that's been happening - selling my RX 480 4GB for $400, then adding 100 more to get a GTX 980 Ti (it's also my first NVidia card since... ever?), which is about 25-30% more powerful than the RX 480.
It also seemed like a good opportunity to find out if blaming my Radeon for lower-than-expected performance was justified or not. Also included the results of a temporary shit-tier GPU (Radeon HD 7770) I used while waiting for the new card to arrive.

Tested on: Core i5-3570K @ 4.3 GHz, 8GB DDR3, Windows 10 Pro 1709.
TF2 configuration: mastercomfig (midquality preset), 1440x900 windowed, mods: explosions, no bullet dust, no soundscapes mod

For any non-nerds out there, I'll also put the performance rating measured in percentage compared to RX 480 (data taken from TechPowerUp - relative performance).

RX 480 4GB (100%): 127.47 fps - https://imgur.com/e0m1umz
HD 7770 1GB (27%): 130,87 fps - https://imgur.com/q8odAc3
GTX 980 Ti 6GB (132%): 130,11 fps - https://imgur.com/AjVVKcM

If you told me this makes no sense, I really wouldn't argue, I don't know what to take away from this. I thought my current CPU isn't bottlenecking my system, but in the other hand, johnf's system with the same CPU (but 100 MHz higher) and a weaker card reaches 200 fps. Maybe comanglia's cfg is just so much faster, idk, I'll try this config tomorrow and update the post.
Or maybe it's the single-channel DDR3 memory... I was planning on getting another 8GB stick to get 16 gigs in total next month, I'll have to test that too.

Looks correct on my side. The one with a smaller font in the menu is on Linux.

The game can still freeze randomly. Rare, but happens. I'll be playing TF2 for quite a bit on Linux, so if I notice anything interesting, I can report here.

So...
I saw the article on Phoronix, showing improvements in the latest Mesa 17 (formerly: Mesa-git) driver for Radeon GPUs. Seeing the huge improvements, I decided to compare how TF2 runs on Linux, compared to Windows.
My first impression on Linux was... that the game runs noticably smoother. I needed some slightly more... quantified evidence for that - and here it is.
Just check this out. Hardware ofc stays the same - Intel Core i5-3350P @ 3.5 GHz, 8GB DDR3 1333MHz, Radeon RX 480 4GB.

Windows 10 Pro x64 (Creators Update, Crimson 17.7.2):

2639 frames 20.047 seconds 131.64 fps ( 7.60 ms/f) 8.975 fps variability

Kubuntu 17.04 (latest Mesa 17 driver from Padoka's PPA, Linux kernel 4.13rc3):

2639 frames 18.673 seconds 141.33 fps ( 7.08 ms/f) 8.006 fps variability

Sure, the benchmarks say only 10 fps of improvement, but the game does feel smoother on Linux. I highly encourage you to try out!

On both OS's I used the same FPS config (Rhapsody's) with the following settings:

rpc_showsbgraph
rpc_enablesprays
rpc_enableshadows
rpc_enablefeatures_low
rpc_disableragdolls
rpc_disablegibs
rpc_experimental_on_amd

mat_bumpmap 1
mat_phong 1

Proof (screenshots):

As promised.

Quick background:

1. Current hardware: Intel Core i5-3350P, 8GB RAM, Radeon R9 280X running on Windows 10
   
2. Replacing the current graphics card with Radeon RX 480
   
3. Tested on two situations. First one: customized comanglia's cfg. Second one: no AA, AF x8, no HDR and V-Sync, all else maxed out/enabled. Shared settings: 1680x1050 resolution, dxlevel 95

First situation, old GPU vs new GPU:

2639 frames 23.632 seconds 111.67 fps ( 8.95 ms/f) 6.561 fps variability

2639 frames 20.597 seconds 128.12 fps ( 7.80 ms/f) 8.144 fps variability

And a second one, old GPU vs new GPU:

2639 frames 30.120 seconds 87.62 fps (11.41 ms/f) 4.648 fps variability

2639 frames 27.595 seconds 95.63 fps (10.46 ms/f) 5.487 fps variability

So yeah, it just confirms what was already mentioned - if you need more frames in TF2, invest on your CPU instead. Getting a Core i7 would surely provide a much higher boost, as shown by previous results on this thread.

I'm waiting for an RX 480 to arrive, which will replace my R9 280X. I'm expecting a good improvement in other games but I'm actually interested in how it does impact TF2, for anyone wondering. Will upload the benchmarks here :)

CPU and overclock: Core i5-3350P @ 3.1 GHz
Graphics Card: Radeon R9 280X

Driver version: Crimson 16.7.3
dxlevel (default is 90): 95
Resolution: 1680x1050
Full-screen or windowed: Full
FPS configs enabled: Comanglia's cfg (w/ facial expressions, gibs and ragdolls enabled; mat_picmip -1), no bullet dust, reduced explosions
Shadows enabled/disabled: disabled

It's either my Radeon being bad at video games, or my CPU somehow not being good enough to keep up the pace...

CPU and overclock: Core i5-3350P @ 3.1 GHz
Graphics Card: Radeon R9 280X

Driver version: Crimson 16.7.3
dxlevel (default is 90): 95
Resolution: 1680x1050
Full-screen or windowed: Full
FPS configs enabled: none (settings: http://i.imgur.com/z7fN5kq.png)
Shadows enabled/disabled: enabled

I'll post one with configs in a while.

After applying all those tweaks and doing some testing, it seems like not much has changed, sadly.
I'd like to keep the game as vanilla as possible, so no extreme FPS configs and "no explosion/bullet/hats" mods. On Windows 8 the game was working fine without those, I can't see why it shouldn't on Windows 10.

Anything else I can try?

Thank you, I'll try those out asap and will let you know about the results. I've already used W10Privacy to do some tweaking (removing apps, disabling some services/scheduled tasks etc) but I'll try out these as well.
Of course I installed GPU drivers from the AMD website.

Any ideas on what would cause that?
On Windows 8.1 the performance was good enough. I was very rarely dropping under 100 fps, and I rarely saw myself dropping to under ~80.
It's worse on Windows 10 though. Even during moderately messy fights I can drop under 40 fps, which is less than favorable.
It was a clean install, not an upgrade. Latest drivers installed. The Xbox Game DVR function is already off.
Specs: Intel Core i5-3350P, Radeon R9 280X, 8 gigs of RAM

That's a shame :( Thank you for trying anyway!

Thank you Hypnotize for spending your free time on making those great HUDs work! I wish I could donate something to you :)

Only one small issue though, with Hudas Iscariote. On matchmaking (casual, didn't check competitive), ranks shown in the scoreboard are very small, quite hard to distinguish. http://imgur.com/a/nMjE5 (Resolution: 1680x1050)
Any way to make them larger? In fact, I wouldn't mind making modifications to the scoreboard, to make the ranks bigger but prevent them from overlapping each other.