The second part of that is more easily answered than the first: because I just do. It's a matter of taste. I like dark fantasy better than I like high fantasy. Of course, the follow-up there is; why not use a system that's designed for dark fantasy, then? It's usually customary at this point to offer up Warhammer Fantasy Roleplaying as the preferred alternative here. There's a few reasons why I actually want it to be D&D, or at least D&D compatible.
1) Familiarity. I know the d20 system quite well at this point, and so do any potential gamers that I'd play with for the foreseeable future. In addition, as I've probably mentioned a few times before, I don't really value system for its own sake anymore. I've more or less made my peace with all the elements of d20 that I used to not like, so now I'm more apt to simply adapt it to my needs rather than look "abroad" so to speak, because I don't get any benefit from getting another system. This kinda gets back to bizarre economics lingo, where economists tried to come up for a unit of measure that could be used to describe intangible benefits. Suffice it to say that I get very few utils out of having a system built according to someone else's conception of what the genre should be and get many more from using a familiar system with a few tweaks to customize it to what I want.
2) It's easier to get people to play D&D with a few houserules than it is pretty much any other system out there. Granted, this is 3.5 with houserules, and it's probably more than just a few, but still... I believe in Ryan Dancey's concept of "network externalities." This is especially true if I decide to run this with my normal group when my turn behind the screen comes up again. D&D 3.5 is our go-to system. We rarely play anything else.
3) It's hard to say goodbye to all the resources I have for d20 that I wouldn't have with any other system. I have, for example, almost two dozen books filled with nothing but monsters compatible with d20, plus many, many other books that have other monsters. If I were to adopt another ruleset, say BRP, Savage Worlds, WHFRPG or any of the other possible ventures, I'd be working in a much more resource poor environment. I have too much stuff I can use in d20, and the opportunity cost of looking beyond that is awfully great.
Now that I've bored you with applying some economic theory to my choice of systems (in bullet point format, no less--I do love my lists), let's get to the meat and potatoes of how to make the game scary, shall we? Here's a few thoughts:
1) Lots of times, I've been advised to make the game deadly, and I have to admit it was a first and obvious thought of my own as well. The more I think about it, the more I think it's a red herring, however. Lots of D&D games are deadly. There are famous meatgrinder campaigns that kill tons of PCs. Some GMs also develop notorious reputations at it. However, no one ever really thinks of horror in terms of these games. In a horror game, I feel that certainly players need to feel that their PCs are at risk of death, at any time, but if they actually show a really high turnover due to death, that doesn't necessarily make the game more horrific, and in fact often has the exact opposite effect; making the game silly, or strategically engaging rather than emotionally horrific. Rather, what makes a game horrific is generated an atmosphere of growing dread. This comes from lack of tension relief, so having too much action, too many opportunities for PCs to die at every turn, relieves tension in the game rather than allowing it to grow and accumulate. Be careful about killing PCs willy-nilly. Allowing it to happen from time to time is often a good thing; make sure your players don't think that you're the kind of GM who'll step in and save them for whatever reason, but other than that, don't go out of your way to make it likely that they'll die either.
2) Related to that, and related to my last post, allow the game to build suspense. Think of an episode of The X-files as a model. The session itself is such an obvious unit of measure in a campaign that you really should use it, plus tension and suspense doesn't carry over well from session to session. See if you can't pace the game to build suspense throughout the session, and peak near the end of the evening (or whatever time that you play), but, again, like The X-files, allow for things to carry over from session to session that contribute to a growing feeling of creeping dread throughout the campaign. The victories of each session don't really represent significant setbacks to the horrific antagonists so much as they do opportunities for you to show the PCs bit by bit exactly how in over their head they've managed to get themselves. If you can engender a feeling at the end of most sessions that they've barely managed to dodge a bullet... again... then the campaign itself will take on an air of desperation and horror. Or, at least, it can.
3) Horror is primarily a mood, and in any oral storytelling tradition, of which tabletop RPGs are a variant of sorts, mood is highly dependent on the storyteller. This means that you need to have moody descriptions. Moody non-player characters. Moody scenes and locations. You need to get your players into the action, and make them feel that they're in a horror story. I used to have linked a thread on a gaming discussion board that was nothing but ideas for disassociated creepy things that you could throw at the players just to make sure that they never rested easy, but you don't need to go that far. Just be sure and describe the game universe in "horror language."