On a recent thread, I promised to compare three Cyrillic layouts according to Carpalx typing effort model. So I measured typing effort for each of the layouts and here are the results: ЙЦУКЕН is worse than DIKTOR, but silghtly better than Rulemak (although I might have messed up something).

Here's what I did:

1) Prepared a corpus of seven Russian books (mostly modern fiction). Carpalx comes with twice as many texts for English, though, and maybe I'll rerun my experiments with a larger amount of different texts.
2) Changed the model parameters, so that ЙЦУКЕН's typing effort on my corpus is 3 and each effort component contributes 1 to the total effort (as described on http://mkweb.bcgsc.ca/carpalx/?model_parameters). It makes comparing layouts easier.
3) Hacked carpalx a little in order to support non-ASCII layouts. DIKTOR, in particular, has a peculiarity: it puts lowercase “ь” and “ъ” on the same key (you have to press Shift to access “ъ”). It's due to the fact that no word in Russian starts with these letters, so their uppercase variants can be safely moved to the top row, making room for more frequently used letters. Carpalx assumed that every lowercase/uppercase letter pair is on the same key, and I had to fix that.
4) Ran carpalx with “reporteffortverybrief” action (I was interested only in typing effort, not in various statistics carpalx had to offer).

Keyboard effort for ЙЦУКЕН
------------------------------------------------------------
k1                      1.480  74.0  74.0
k1,k2                   1.888  20.4  94.4
k1,k2,k3                2.000   5.6 100.0
b                       1.000  33.3  33.3
p                       1.000  33.3 200.0
ph                      0.000   0.0   0.0
pr                      0.476  47.6  47.6
pf                      0.479  47.9  95.5
s                       1.000  33.3 100.0
all                     3.000 100.0 100.0

Keyboard effort for DIKTOR
------------------------------------------------------------
k1                      1.450  83.6  83.6
k1,k2                   1.696  14.2  97.8
k1,k2,k3                1.734   2.2 100.0
b                       0.558  21.9  21.9
p                       1.176  46.1 147.5
ph                      0.000   0.0   0.0
pr                      0.260  22.1  22.1
pf                      0.873  74.2  96.4
s                       0.816  32.0 100.0
all                     2.550 100.0 100.0

Keyboard effort for Rulemak
------------------------------------------------------------
k1                      1.978  83.4  83.4
k1,k2                   2.328  14.8  98.1
k1,k2,k3                2.373   1.9 100.0
b                       0.509  15.7  15.7
p                       1.864  57.7 127.3
ph                      0.000   0.0   0.0
pr                      0.296  15.9  15.9
pf                      1.458  78.2  94.1
s                       0.858  26.6 100.0
all                     3.230 100.0 100.0

Basically, Carpalxt takes each triad in the text (a triad is a sequence of 3 consecutive letters; they can overlap), computes typing effort for the triad and sums individual efforts. The interesting parameters are:

* b — the finger travel distance for a triad;
* p — the penalty for using a letter from an inconvenient row, or for one that is pressed with a weak finger like pinky;
* s — the stroke path effort. Basically, it tries to penalize “uncomfortable” triads (no alternation for hands, rows or fingers).

The sum of this values is the final score for a layout (the lower the better). I didn't analyze the results thoroughly, but is seems like Rulemak is better that ЙЦУКЕН in every aspect except for finger penalties (pf). I'll try to find out what causes that.

Any questions or suggestions are welcome.