So essentially, the problem with Paul Almond's suggestion is that one can find languages X that are 'X[0]-pathological' in the sense that almost all of their children are both (i) good at implementing some fixed language X[0] and (ii) even more X[0]-pathological than they are (this being a recursive definition of 'pathological').

Such languages destroy any hope that Paul's "low-level index for Y" won't depend on which language X we start from.