Could you paste some of these (and better yet, minimal corpi that generate them).

The only such chars I can think of would be escaped ones like \n \t \\.

Nice analysis of yours, Dreymar :-)  And yes, Colemak is a nice allround layout :-)

After many experiments I came up with this layout.

Hands: 56% 43%   Fingers: 9.0% 11% 21% 16% 0.00% 0.00% 14% 12% 10% 8.0%
:  U  O  P  Y   X  M  L  B  V
A  I  E  N  H   G  D  R  T  S
<  >  ?  K  Q   F  C  W  J  Z

,  u  o  p  y   x  m  l  b  v
a  i  e  n  h   g  d  r  t  s
;  .  /  k  q   f  c  w  j  z

Fitness: 71426   Distance: 59945
Finger work: 11  Inward rolls: 7.60%  Outward rolls: 2.75% 
Same hand: 35.53%  Same finger: 1.06% 
Row change:  11.11%  Home jump:  0.56%  Ring jump: 1.36%  To center:  4.25%

How did I design it?
After trial & error these steps worked well:
1. Find text to analyse: your own reports, code, email; books; websites;  in the language(s) you use. My input was 80% Dutch, 15% English and the rest was German, Spanish and some French.
2. Paste it into an editor. I used Medit, but many text editors will do.
3. Get rid of empty lines and line breaks (use the regex  \n  \t  and so on).  LibreOffice is lousy at this, I found out. An editor does it many times faster. Save it as a .txt file.  Encoding doesn't matter
4. Analyse the file using lalops Python program. Use -n1 for the outputfile allChars.txt and -n2 for the outputfile allDigraphs.txt
5. Clean the two output files. There will still be unwanted characters in the .txt files, from page-breaks, code from tables etc. I found that the humble nano editor is best for this (wow!). My builds are in folder Builds, so I type:  nano ~\Builds\Typing-master\Typing-master\data\allDigraphs.txt  and nano opens that textfile. Next I look for unwanted stuff, which is mostly stuff like \t\t and so on.  ctrl-W \  takes you to the next lines that contains a  \    ctrl-K removes that line.   (yes, a simple python progam could remove all these things as well, but I'm still learning python.... )
6. Make the layouts with mtgaps program (you have to build it first on your machine). Optionally: experiment with different values for inRoll, sameHand etc.
7. Check the layouts in patorjk's website. Feed it with chunks of text to see how your layout looks & performs.

How does it perform ?
According to patorjk's website:
- it scores the best for Dutch texts.It clearly beats Colemak, Dvorak, Balance 12 and so on.
- It beats Colemak in English, German, french, Italian and Spanish as well. Often it ranks the best, other times as second or third after Balance 12 or HIEAMTSRN. But always before Colemak and Dvorak.
- I tried some other variants, some more " rolling"  others more " alternating", which scored good as well, but not as good as aienhgdrts.

Conclusions
- Yes, Colemak is still a very good layout, and has all the advantages. it get picked up by the market and so on.
- There are some other layouts that are very good as well: Dvorak, Balance 12, the standard (English) MTGAP layouts
- if "qwerty similarity"  is an advantage, i don't know. In spoken languages I confuse languages that are too similar - I can't easiliy switch from French to Spanish, for instance. To German is easier.  Back to keyboards: I don't like one hand key combo's like ctrl-X and ctrl-v because it gives me a sore hand. If that is important to you, Colemak may be a good choice. But to me it isn't

- Making a dedicated layout is worthwile (on paper - I have to try it out in real life !) but is of course more of a hassle. On the other hand: if you are going to learn a new layout, why not go all the way?

Yes, you are right: it's part of the corpus, But... MTGAP's program complained about escaped characters etc, it refused to run until I removed those. Furthermore, I didn't calculate the location of the tab, the return keys, the space, so I felt free to remove it.

BTW, my 'dream' program would be a combination of your Python analyser + MTGAPs layout calculator + patorjk's layout visualiser +  mikekuehn's "try keyboards in the browser".  Imagine: feeding the program raw text and seeing it return a layout, visualized, that you can tweak and immediately try in the browser.....  ?  Or even better: also have it generate a Windows PLK file and Linux XKB files..... :-)  'Why don't you make it yourself?' one might say... Good question! Well, because I can't program at all :-(  That's why I am grateful for the work of you guys who can and do that! :-)

Update. I've done some more research, bigger corpus. I come up with a slightly different but better one:

Hands: 57% 42%  Fingers: 9.0% 10% 22% 16% 0.00% 0.00% 14% 12% 10% 8.0%    There is some imbalance left/right, especiallu the left middle finger has a lot to do. 

/  U  O  P  Y   X  C  L  B  V
A  I  E  N  H   M  D  R  T  S
;  <  >  K  Q   F  G  W  J  Z

.  u  o  p  y   x  c  l  b  v
a  i  e  n  h   m  d  r  t  s
:  ,  ?  k  q   f  g  w  j  z

Inward rolls:  7.54%; Outward rolls: 2.94%; Same hand:     36.39%;  Same finger:   1.46%; Row change:    12.27%
Home jump:     0.66%; Ring jump:     1.44%; To center:     3.94%

I call this one Juli16

Patorjk scores this one much better than the previous one.

The remarkable thing is: I compared in (in Patorjk) to the layouts Colemak, Dvorak and Balance 12. For various languages. These are the winners. 

For Italian: winner is Juli16
For German: winner is Juli16
For French:winner is  Juli16
For Dutch: winner is Juli16
For spanish: winner is Balance12,  # 2= Juli16 #3/#4 = Dvorak or Colemak
For English: winner is Balance12,  # 2= Juli16 #3/#4 = Dvorak or Colemak

So there you have it. for the record: all these layouts are very good; Colemak, Dvorak, Balance12 and Juli16.

Now it's time for live testing :-) Edit: I will also implement some other ideas. Such as:
- making the key of q type qu   Because in most languages q is always followed by an u.   
- always having a space after ,
- making the modifiers sticky
- always having space and Shift after .
- working with layers (thank Dreymar!). In the numerical layer, the above rules do NOT count, so I can type 1,23  or xq+z or 10.000