I don't know the answer for certain, but am willing to hazard a guess that the most efficient way of transferring between a circular orbit and an elliptical orbit, assuming for a moment that they are coplanar, remains a single Hohmann transfer.
Here's why: Nominally, a Hohmann transfer takes you from one circular orbit to another. For every point on an elliptical orbit, there exists a circular orbit which intersects that point with the body moving in the same direction (though not at the same velocity). A Hohmann transfer works by first turning your orbit from circular to elliptical, then back into a circular orbit.
Changing the eccentricity of an orbit requires a burn delivering the proper velocity change (including the burn attitude) at the appropriate point in the orbit. This is done at the end of a traditional Hohmann transfer orbit to re-circularize the orbit at the desired distance from the barycenter, which becomes our new orbital radius. Once our orbit is circular, we could, in principle, apply another burn to change to a different elliptical orbit with the semi-major axis being the circular orbit's radius. By allowing the time between the burns to go to 0 and recalculating the second de-circularization burn based on the new point in the orbit at which it happens, we can turn these two burns into a single burn.
Thus, by tuning the Hohmann re-circularization impulse, it should be possible to transform your Hohmann transfer orbit directly into an elliptical orbit of the desired shape, while maintaining the Hohmann transfer's (delta-v) efficiency in getting between the two corresponding circular orbits.
It follows from the same line of reasoning that this should work equally well whether none, either or both of the endpoint orbits involved are circular or elliptic. (If the origin orbit is elliptical, you simply, at some point during the first burn, are instantaneously in a circular orbit.)