I doubt there's a pretty answer for this; Elo difference only gives a rough estimate of the probabilities of a win on one side.
Instead of computing this analytically, you could estimate it empirically, either by looking at the length of relevant games in a large database, or by simulating a large number of games with engines at the appropriate strength settings and collecting statistics. But again the answer you get may not be particularly meaningful.
I was wondering if there was a formula for calculating an expected no of moves to win between two players of significant difference in rating? Outside my mathematical abilities, but think it may have too many parameters effecting to get a meaningful answer.