Ok, I did some tests, as the_foz_4 proposed, I used exactly the same conditions as was his proposal (including morale off). Difficulty was on normal.

I used Dismounted Feudals (charge 0 , shield 0) and Dismounted Normans (charge 0, Shield 6, defence skill + 6)

I ran 10 battles. I controlled Normans at first.
Results: They won in 9 battles from 10! The remaining number of soldiers:5,16,11,35,7,3,7,31,22. In the battle, where Feudals won, 17 of them remained.

One thing I noticed was the fact the computers soldiers enter combat differently from mine soldiers. Computers unit attacked "from move" - they moved and suddenly attacked. My unit moved, then halted in close proximity of enemy and then charged. So this is something we must calculated with.

Then I ran another ten battles, but controlled Feudals, nothing changed.
Results: They won in 9 battles from 10! (what a surprise...) The remaining number of soldiers: 16,27,18,11,18,8,22,6,7. In the one battle, when Normans won, 7 of them remained.

Conclusion:

In the first set of battles, 137 soldiers remained on victorious side, in the second, 133. I think the results are pretty similar, so the shield probably really give penalty to defence as high as is its number.

But then we have the massive discrepancy between win/loss ratio of human vs. computer controlled unit with supposedly similar stats...

I will do another set of tests, this time with units using vanilla stats.

But it seems that even the computer ability to control units is bad.