I guess it really depends on the kind of game you're going to make. With content-driven games, the off-the-shelf solution is a good one. With new types of gameplay, you might want to go to the custom engine.

I also agree it depends on the kind of game you're going to make. I have limited experience with open source engines, but from that experience, I can see how they can be effective and efficient tools for your project. If the engine is built such that its functionality COMPETENTLY provides the fundamental framework for your game, indeed it SHOULD save you at least a decent amount of development time (if not then you've wasted time/money). Considering this, I think its easy to see why some publishers are willing to spend the money to license them from outside their doors.

That being said, I very much agree with Lesley Matheson in that unless you're developing a very uniform game, you're likely hard-pressed to not have to make some tweaks to properly implement some of the features of your game.

As a learning game programmer, myself, I currently have (and I can't see it ever changing) a preference for building my own engine(s) rather then opting for something open source. In fact, when making a few pedagogical 2D games, I developed a simple, yet easy-to-use and effective engine that was designed to meet the needs of my endeavors for the foreseeable future and is easily upgradeable. The design was so effective that I made 3 complete games including a multi-level, boss battle-featured Space Invaders game on it within the span of a little over a month.

And I just wanna say this last thing. I think probably one of the biggest resulting benefits from building your own engine, especially for the programmers, is that when it comes to executing your products on them, you have a much more intimate understanding of what is going on under the hood than if you were to borrow somewhere else's middleware and learn how to use it.

This strikes me as bizarrely bad advice, in an age where the role of middleware is so commonly recognized as an essential part of building games. Although I can agree with the previous posters who rightfully point out that the usefulness of middleware does depend on the kind of game you're trying to build.

But in general if you're building a complex game that uses many standard game components, such as physics, sound, 3D graphics, a user interface, keyboard or gamepad input, networking, asset exporters, and perhaps some standard AI like pathing, you'd put yourself back years of dev time trying to replicate engine technology which is already readily available, tested, and stable.

In my opinion, it is far better to decide what it is about your game that is going to be different, focus your technology effort there, and fill in everything else with middleware. Contrary to what this article suggests, I find that inexperienced engineers tend to want to write everything from scratch themselves, and experienced vets recognize the advantages of reusing existing technology.

Unless you are building an FPS derivative (reskinning the other cliche bullshit) then you are not going to get a good solution from epic or crytek or the others. Epic sells their shit as an end all be all and it simply isn't. It ends up requiring massive amounts of custom work anyway if you are doing anything out of the box creative, so you might as well build your own tech in that case. If you are building an FPS then awesome. Use what the others have. However the prices for the big engines are disgusting right now. The royalty demands are even worse. Also, they are full of lies when it comes to what their salesmen are pitching, so beware. Do your homework. If you have mediocre engineers, the safe bet is middleware. If you have full budgets then owning your own tech is a better bet as it increases your company value tremendously. Look at it this way. The truly independent companies almost all own their own tech. You think Epic would be worth what it is without the tech? Not even close. If you are building long term value in your company and not just looking for the quick solution, building tech is the way to go.

I see the advantage of building your own engine - and not doing so. I have recently worked with UE3 which helped the production focussed on creating a game instead of a technology. With sparse amounts of programmers around, buying an engine saved us some ressources in the end. Sure it costs money, but making such an engine hadn't been within our grasp anyways.
But if/when settling with an engine, it goes that you have to work WITH the engine - not against it.