Marshall first played it back in 1918 against Capablanca. He had been analyzing the defense for years and came to the conclusion that it was sound...not that it wins, but that it gave black winning chances and should be good for at least a draw. Marshall actually lost his game against Capablanca who accepted the P and defeated Marshall's attack even though he was facing a prepared variation he had never seen before.
I am not sure what to make of the Marshall. It's still a sharp one and to play it requires a lot of theoretical knowledge. That's why many players prefer to avoid it with moves like 8.d4, 8.a4 or 8.h3 instead of 8.c3.
GM Greg Serper wrote that the days when the Marshall Attack was feared are long gone. He added that these days it is a weapon of choice for many 2700-plus GMs looking for a quick draw. Using their computers, they have analyzed the Marshall Attack to an almost forced draw. On the other hand, some players still search for new ideas or even old, forgotten ones. As a result, in OTB play black's attack, which is based on positional considerations rather than tactical ones, can be dangerous. The opening has been analyzed in some positions beyond move 30!!
I have used the Marshall only twice in correspondence games, losing one and drawing one. In the game I lost my opponent sent me a message when the game was over saying, "Always happy to play against the Marshall!" I faced it once with white and drew.
My ChessBase database shows white wins 28.6 percent, loses 22.8 percent and draws 48.6 percent. The Chess Assistant database percentages give a different picture: +39 -34 =27
In the article Marshall's analysis was mainly concerned with 18.Re3 and 18.Bd2. He gave the main line as 18.Re3 and gave 18.Bd2 scant attention adding that he was intuitively convinced that it was inadequate. And that's what I wanted to investigate...how good was Marshall's intuition?
His mainline was 18.Re3 which Stockfish immediately assigned a losing evaluation. As for 18.Bd2, it was the best and was immediately the top choice, so it appears Marshall's intuition failed him this time.
The engines indicate that not only was his intuition wrong (in this case), but his analysis was really not very good either. Analysis is always a problem with old books and magazine articles because sometimes they were based on results...every move by the winner was praised and the loser was often depicted as a hapless fellow who was badly outclassed. In fact, many times the games were tough struggles where the annotator missed (or just plain ignored) better moves. Also, much of the analysis was superficial due to it being rushed. Few players had the temerity to question a master in those days anyway! Then, of course, they didn't have chess engines that immediately point out hidden tactical resources that it might take a human a long time to discover. Then, too, I wonder if sometimes masters didn't publish faulty opening analysis just to catch an unwary opponent.
All GMs warn us not to always trust small pluses and minuses given by engines and human insight for proper evaluation of positions is required. Naturally, that's the problem for most of us...we lack the ability to correctly assess a position. As Wolff Morrow pointed out in his article referred to in a previous post, it is still possible for a 1500 player to reach a fairly high level in correspondence play IF he is experienced in properly using an engine. When it comes to playing over published games, especially old ones, most of us aren't going to spend hours and hours letting an engine examine a position and doing a lot of interactive analysis. If there is a question about the accuracy of an engine evaluation of a position, then I like to use the Fritz Shootout function which plays out the game at several different plies. By using this method you can get an idea of the direction the game is headed and by looking at the completed games you can see how the engine produced the results it did. One word of warning though...I've seen some of these go on for well over a hundred moves and in practical play that leaves a lot of moves where humans can go astray.
So, here's the results of the analysis of Marshall's article using Stockfish 6, Komodo 8 and Shootouts conducted a several places.