As I see it, the key issue surrounding Comic Book Science is the amount of detail included in a description. The appropriate amount of detail depends on several factors: Setting, Author Knowledge, and Audience. I'll tackle each of these one at a time.
The most important element to consider is the setting. When writing for a setting far removed from real life, the author has much more room to invent technology. In a science fiction game where space travel is common place, author's have a lot of leaway. In other words, if they want to claim the Transmogitive Incipator is the core of any Anthosian Propulsion Unit, nobody can dispute it. But when the setting is closer to real life, authors must be careful. For example, you can't claim the laser in your air conditioner malfunctioned causing the VCR to bake your mail. It makes no sense, and everybody knows why.
Author's should avoid writing about things they know nothing about. That's just common sense. This doesn't mean an author can't research a subject. The danger is an author may only think he knows what he's talking about. Writing about subject close to your professional background is probably the best way to avoid this problem. But remember, there's always somebody who knows more about a subject than you. Avoid calling yourself an expert, because people these days are more likely to look for flaws if you do.
There is little need to describe how every little detail of an ultra-modern, high-tech device because the players and the GM using the device are only going to be concerned with what the gadget does. In some cases, they'll worry about how the gadget does what it does, but this will most likely be abstracted by the skill system. This is probably the most important point I want to make: concentrate on the capabilities of Comic Book Science, not the implementation.
Let me present an extended example. It is common place for authors to want to justify why a computer is actually artificially intelligent - an AI. Some go on to say the computer can actually change its programming, or it is self-programming. What a unique concept, or so they believe. Self modifying code has been around for decades. It is not difficult to write code like this. And, just because you can do it, doesn't mean its helpful. This is the case with self modifying code. A better way to justify the AI would be to claim it was developed by a genious. As far as the story in concerned, it doesn't matter how the AI works. The only thing which matters is what the AI can do. If, during the course of a game, somebody wants to figure out how the AI works, it will all hinge on the character's computer knowledge. In other words, the skill system will take care of it. The character makes a computer programming roll, and then he knows how it works. If the character wants to change anything, again he'll either have to do it through the skill system (by making a roll) or through role-play, which will be done at a high level. In other words, during role-play the character will not be asked to actually write the new program. The character will describe the changes, and the GM will decide what happens, possibly based on the results of a skill roll.
This advice can be extended to areas other than science. Consider "Comic Book Government" or "Comic Book Law." These areas also tend to be abstracted. This is because they can be complex subjects. These subjects don't fall within the common knowledge of most of us, so they get twisted to match our conceptions. Just because you know what's right and wrong, doesn't mean you know what's legal and what's illegal. Again, concentrate on what needs done, and not on how it is accomplished.
I suppose I can sum up Comic Book Science in one line: Less (detail) is more (believable).