I find at my level that the database numbers are virtually meaningless. As long as the opening isn't just idiotic, and it suits my style, I'll play it.
Regarding the Caro-Kann v. Sicillian, unless you and your oppnent both play flawlessly for the first 20 moves or so, those numbers really aren't applicible. Go with the CK (or the French) if you like closed positions.
Database statistics mean very little. Don't play a move when it's actually a mistake, an actual bad move (slightly worse is no problem).
What works best for you is much more important.
I agree statistics are useful in determining what should happen with best play. However since I doubt you'll be playing Shirov, Topalov or any other top level GM anytime soon, don't put too much stake into them.
I would suggest that since chess is a hobby why don't you make decisions that maximize the enjoyment of your hobby? If you prefer closed positions, play the Caro. Now if you simply don't grasp this defense and just lose tons of games with it, then maybe you should look elsewhere.
When deciding what openings to play, how heavily do you weight what is "objectively" good against what suits your individual playing style? As an example, I feel as if I should study certain lines of the Sicilian defense because the databases I have seen show that they offer black the best odds against 1. e4. I also know that I tend to play best in very closed positions, and sometimes wonder if my time might be better spent studying something like the Caro-Kann defense, even if statistically it is slightly less viable than the Sicilian.
Your thoughts? How much should I pay attention to statistics, anyway?