Here's a reply to this topic that I originally posted in the DragWars mailing group:
It simply gives a different probability curve, which may be more appropriate for the type of check being made (I've not gone through the rules to see if this is actually consistent). I believe common checks, or those where randomness is desired, tend toward a D20 roll. Infrequent checks, or where too much randomness can be frustrating, use the 2d10 method.
For combat, a D20 provides a nice flat chance of getting a 1 to 20, which results in a greater degree of randomness in combat. This is appropriate, as combat rolls occur frequently during a typical encounter, and the player has both Armour Factor and Health Points that act as a buffer before unrecoverable failure (death, usually).
On the other hand, Stealth checks use 2d10 which are rolled much less frequently per 'Stealth' encounter, and a failure tends to be immediate - there's no equivalent of Health Points to take the brunt of a few failed rolls. The same applies to Evasion, where the penalty for missing the roll can be far more severe than a bad combat result.
You can use 2d10 for combat if required, but note it will make combat last longer between equal foes, and small differences in skill will be amplified. To illustrate:
Let us assume 2 basic humans, Attack 11 Defence 5, squaring off for a fight.
With a 1D20, they each have a 30% chance of scoring a hit on one another per combat round.
With a 2D10, they each have a 15% chance of scoring a hit on one another per combat round.
Furthermore lets assume one of the humans is a bit slow, and has an attack of 9 instead of 11.
With a 1D20, Attack 9 human has a 20% chance of scoring a hit.
With a 2D10, Attack 9 human has a mere 6% chance of scoring a hit (hardly more than the chance of a critical in the usual rules).
On the flipside, lets assume one of the humans is a high ranking character instead, with an attack of 20.
With a 1D20, Attack 20 human has a 70% chance of scoring a hit.
With a 2D10, Attack 20 human has a 79% chance of scoring a hit.
Basically, when using 2d10, every point greater or lower than a target value of 11 logarithmically reduces or increases the chance to hit. Since the standard difference between attack and defence is 6, there's already a pretty heavy bias toward a very low chance of success.
The combat for DW seems to have been balanced around the premise that a normal being should hit an equivalent opponent about 30% of the time in a combat. To keep a similar ratio when moving to 2D10, you would need to add 1 point to the Attack score and deduct 1 point from the Defence score of all creatures and professions listed in the rulebooks - this would change the standard difference between attack and defence to 8, which give a 28% chance to hit. Pretty close to the regular 30%. Note this already happens to be the standard difference between Stealth & Perception, which uses 2D10 for checks.