Shrinking range of openings

This recent comment by GM David Marciano caught my eye: 'Modern chess is very different from the chess of the 1990s, and even more so than the chess of preceding decades. The exponential growth in the strength of computers has effectively transformed the work of opening preparation into computer aided scientific research, and the range of playable variations doesn't stop shrinking as the years pass ("ne cesse de se rétrécir au fil des années").' [Europe Echecs, April 2006, p.15] I'm fairly certain that my translation is correct, except perhaps for the last clause, which is exactly the part that caught my eye.

I know that opening theory is advancing steadily and that many opening variations (Marshall Gambit, Najdorf Poison Pawn) are analyzed well into the middlegame and even the endgame. I also know that a few popular openings of yesteryear have almost been refuted (Benoni). This is certainly a negative trend for chess.

I was also under the impression that computer aided opening research (CAOR?) is helping to open up promising areas of play in variations that have never been seriously investigated. This includes just about any reasonable move that has never been played at the GM level. This should offset the number of lines which have received the most CAOR.

The largest chess databases only contain about 10.000.000 (10^7) games covering maximum 10^9 positions, the majority of them played by amateurs. With chess having over 10^100 postions, and even taking into account that many of these are clear wins for one side or the other, that still leaves many positions which have not been, and may never be, investigated.

Is GM Marciano being overly pessimistic, or am I being overly optimistic? If I'm being too optimistic, perhaps it's time to take Chess960 more seriously.