Personally, I see the "dread companion" issue as primarily a character-development device for the PC-familiar/mount relationship. Too often in D&D, this relationship is neglected or ignored, with the PC's animal being treated like a pet or, worse, a mobile magic item (i.e. tool). In a setting as character-driven as Ravenloft, I think it can acutally be a *good* thing that PCs are motivated to take more responsibility for their familiars' and mounts' lives; we DMs just need to be gracious about rewarding sincere PC efforts to discipline and teach their faithful (though evil) companions how they ought to behave.
Instead of assuming that dread companions are actively malign, I think it works better to have a new familiar/mount lack any pre-existing concept of good or evil, and choose its methods solely for their expediency. If it's placed in a position where the quickest, surest means of accomplishing a task is to murder, steal, betray, or torture, it'll do so without a second thought; when first summoned, it has no clue that such methods might be objected to by its master, let alone that they're morally wrong! Once the PC discovers its misdeeds, it's up to the master to chastise the dread companion for its behavior, and lay down the law that it mustn't do such things again. This may have to be repeated several times, as the dread companion struggles to distinguish what's acceptable from what isn't (e.g. if it kills in combat or self-defense, that's different from killing because an innocent bystander was in its way), and the PC strives to explain morality to a creature which, in fact, *can't* learn the difference between Good and Evil, and will never understand why hurting others is even an issue.
While turning a dread companion non-Evil isn't possible, I'd allow a well-disciplined familiar or mount to eventually cease ACTING Evil ... if only for the sake of its master's feelings on the subject. Mounts and familiars are intelligent and able to learn, particularly as their Int scores rise, and they do *want* to please their masters. So if a player makes the effort to role-play the companion's "education", it'll reach a stage where it grasps the limitations of its master's moral code (even if it's baffled why the PC would NEED such a pointless thing), and no longer commits the offenses it once resorted to as convenient. By this stage in a campaign, the familiar/mount will have developed an individual personality through role-playing, and no longer needs to be monitored for ethical lapses to maintain its identity as an NPC. Of course, it may still commit evil acts if faced with a dilemma it hasn't yet been taught to deal with; Ravenloft has a way of presenting unique ethical challenges, often when they're least expected.
The familiar of an Evil-aligned spellcaster, of course, would start out doing evil deeds for convenience, but would become more corrupt over time if its master endorses such behavior. Even as the dread companions of non-Evil masters become more "civilized" in their behavior, those of the villains eventually turn as willfully vicious and cruel as any darklord, as they learn to *enjoy* actions that once were merely pragmatic.
