Ruy Lopez, Marshall Attack, Steiner Variation!?
There are two opposing schools of thought as to how intermediate players should approach the opening:
Specialization - the specialization philosophy recommends that one specialize exclusively in a White opening (e.g. e4 with Italian, c3 Sicilian, Tarrasch French, Panov-Botvinnik etc.), and a Black opening against e4, and one against d4, for several years. This is supposed to give you in-depth experience in your opening and will probably bring better results in the short run.
Broadening - the broadening philosophy recommends that you learn and play a range of openings both with White and Black, over a few years. This is supposed to give you a broad range of experience with different ideas in the opening, resulting pawn structures, and typical middlegame plans, and be good to your overall development as a chess player in the long run.
Like with all diametrically opposed views, the best approach is probably somewhere in-between. I've solved this for myself by currently going to the specialization route for live play, but playing lots of different opening-specific daily tournaments.
This brings me to a recent daily tournament on the Marshall Attack, which I won (see here). For playing the white side I used the help of two books, Bogdan Lalic's "Marshall Attack" and David Vigorito's "Understanding the Marshall Attack", trying both the d4 and d3 lines.
However, as Black, I played the rare Steiner Variation (9...e4), with great success, winning all games.
I first heard of this variation in GSerper's excellent article: Openings for Tactical Players, Ruy Lopez Marshall Attack.
Of course, theoretically it is considered to be defanged. Lalic devotes a whole chapter on it, prefacing it by saying: "...with best play White retains a clear advantage." (p. 104) (Vigorito only has the following to say "The old Steiner Variation, 9...e4, has been considered dubious for years and won't be considered here." (p. 7)) (To give you some idea of the frequency of its occurrence at master level, Opening Explorer has 1436 main line games and only 97 Steiner games).
However, in this case, the best move is not particularly natural and one has to probably know it beforehand. In fact, GSerper shows a number of games where surprised GMs play natural moves like 10. Ng5 or 10. dxc6, exf3, 11. Qxf3, and get wiped off the board fast. Let's look at some of the theory:
Natural move 1: 10. Ng5?!
Best Play: 10. dxc6, exf6 11. d4!
I hope that this has convinced you to give the Steiner variation a try.