Aliyah, when using a computer to analyse a position, human interaction is very important to get a good result. 

Set up the position. Let the computer think for a minute or two. Look at the evaluation but not necessarily the moves suggested. Play some reasonable moves for each side. You will often notice that the evaluation changes when you do that. The horizon of the computer calculation is moved forward. Go back and test other moves. Be sure to test moves you suggest yourself and not only moves suggested by the computer. Also, take notice of moves that changes the character of the position and the evaluation of them.

Post the games here and I'm sure we can find improvements for white.

Yes. Gambit play lives! Raging Rudolf would rejoice.

Way to go Alias!

You succinctly pointed out who to trust and why, not only in this position, but in many other cases in which the computer wins a pawn and claims an advantage without understanding the compensation.

Aliyah -

Alias is correct.  You'll get into trouble with 8...Qxg2.  Better to play 8...Na6.

- Lost Highway

This is the name of a variation that appears in Sam Collins' recent book, Understanding the Chess Openings:

1 d4 d5 2 c4 e6 3 Nc3 c6

It is called the Triangle in Glenn Flear's recent book on the Slav and Semi-Slav.

The Marshall Gambit occurs after:

4 e4 dxe4 5 Nxe4 Bb4+ 6 Bd2 Qxd4 7 Bxb4 Qxe4+ 8 Be2

Now all the books and databases agree that Black's correct response is 8...Na6.

However, all the engines (Fritz7, Fruit 2.1, Pro Deo) claim that Black is at least half a pawn better after 8...Qxg2, and indeed it is hard to see why Black can't successfully defend, even after 9 Bf3 (...Qg5) or 9 Qd6 (...Ne7).

Whom should I trust?

Thanks for your ideas.