I agree with @jank's answer. There's really nothing wrong with practice / order effects per se, and, in fact, the detection of these order effects can sometimes represent theoretically interesting contributions in their own right.
It is true that, without counterbalancing and when tasks A and B represent the within-subjects manipulation of interest, the order in which tasks A and B were presented becomes confounded with condition making inferences about the manipulation of interest ambiguous. However, in some cases, for example when tasks A and B are two measures and the primary focus is on a between-subjects manipulation, even this is not necessarily a concern.
If you're really concerned about order effects, the safest way to approach the administration of two tasks is to do it in entirely separate sessions. This is something my lab does routinely on tasks where we think taking one will influence future tasks. Of course, this sort of design is not always possible.
In short, what you do with respect to order effects really depends on the specifics of the study design. Having order effects of unequal magnitude should not be a concern in the slightest, especially if you have designed your study in such a way as to allow you to measure their magnitude.