There are many roleplaying games - by which I mean published series of books or equivalent, rather than instances of those games - differing in purpose, emphasis and quality. Typically a game consists of a setting and a system, since fundamentally an RPG is part rules-competition and part storytelling; some products are all system (because the publisher is going to bring out multiple settings or because they expect you to write your own), or all setting (such as the many 'campaign settings' produced for D&D and other games). Today I'm talking system, because a conversation at one of the forums I haunt has reminded me of similar arguments in the past.

As I've grown older and 'wiser', played more games and done more work on designing settings and systems, I've noticed three distinct stages of opinion on systems. I'm not claiming that they're a necessary progression, or that each is better than the last (although I thought so, otherwise I wouldn't hold them), but they're the positions I've had over the years.
When I first started playing, and more to the point when I first started running games, I followed the rules of the system, whether I liked them or not. After all, someone much wiser than me had written them, so they must be right.

Before long it's clear that although someone wise wrote the system, he didn't write it for my game. If the rules don't suit my game, I can change them.
That's all well and good, for the right game. The more technical the system is, the more value there was to the playtesting that the publishers [hopefully] did, and the easier you'll break the whole by changing some part, but that's still your right. A rules-light game in a rules-light system can gain a lot when the ref is comfortable brushing over gaps in the system with a quick ruling, and making those judgements is one of the important skills of running a game.

What I have noticed in some people is a blasé belief that since the ref can change the system at will, it doesn't matter what system it is. I probably sounded like that myself at some point, and a lot of players are proud of having reached the second stage and only too happy to spout this hearth wisdom as though it's a useful statement.
The third stage is the realisation that although the ref can change the system, he shouldn't have to. So a character does something not covered by the rules, something really obscure: fine, the ref makes a judgement call - picks some skill or stat to make a roll against - and everyone forgets it. If this really obscure action turns out not to be so obscure after all, the ref needs to try and be consistent in his judgement, and soon a full-blown house rule is born, and needs to be recorded or remembered. Sooner or later that can be a lot of work, and it would be work saved by using a system that already covered the eventuality.

A while back I looked at the d20 system - the system behind the most recent D&D among others - because I wanted to base a game on it. I'd chosen the d20 system rather than starting from scratch because there's an open-source style license (albeit a hideously viral one) that allows derivative works to be published, so it looked like a good head start.
The further I got into the project, the more work it was. Although there were a couple of nice features I wanted from d20, I didn't want most of the combat rules and I didn't want classes or levels, so I had to redesign everything to get rid of them. That was a heck of a lot of work, and I soon realised that actually it would be quicker to start from scratch and work from the other end.

So no, I'm probably not going to run D&D again, despite all the books I've amassed for it over the years. (Although the books are always reusable: setting can be reused wholesale if you like it enough, and often nice pieces of system can be stolen and converted). And that's a shame, because there are some real gems in that system, but once you chip away all the crap there simply isn't enough left to run a game with.