The test program has one interface ICar:

interface ICar {}

and two implementation classes of ICar:

class Ferrari : ICar {
    private string name;

    public Ferrari(string n)
    {
        name = n;
    }
}

class Bmw : ICar {
    private string name;

    public Bmw(string n)
    {
        name = n;
    }
}

First of all, four objects are stored in the database:

Ferrari: name = "F430"
Ferrari: name = "F450"
Bmw: name = "Serie 5"
Bmw: name = "Serie 7"

Then the following SODA query returns different sets of objects depending on whether the interface ICar has one or more than one implementation classes:

IQuery query = db.Query();
query.Constrain(typeof(ICar));
query.Descend("name").Constrain("F450");
query.Execute();

The query returns
    * all the four instances of ICar, if Ferrari and Bmw both implement ICar;
    * the Ferrari object with name="F450" if only Ferrari implements ICar.

I checked out the code in QQueryBase.AddInterfaceConstraint, and found out that if I replace

constr = constr.Or(Constrain(yapClass.ClassReflector()));

with

constr = Constrain(yapClass.ClassReflector());

then I get the correct query result.