Yeah, me and my buddies used to make RPGs all the time. Mostly, we'd just think up "awesome" ideas and make up our own rules, because porting rules over was apparently "lazy". Most of the time, our made up rules kind of sucked.
Later, in high school and whatnot, it got a bit better. Then, it was just sort of house-ruling the bajeezus out of a system to get what was desired.
In 2003 or so, I made "the mutant game", which was initially based off the d20 Omega World mini-game but then stripped down of all silliness and made into a more D&D-like game. It was a lot of fun, and it's gone through numerous versions.
I also wrote up the "d12 system", which is sort of a fast-running game that worked pretty well - the problem with it was that character experience wasn't handled very well, with only five "levels" of ability for a character (sort of like Savage Worlds).
Currently, in between 4e games and in those rare moments I have free time, I'll fiddle with an unnamed percentile system. I don't really have it tied to a setting yet, but I know it'll be modern or futuristic, and probably either space opera-y (sort of like Mass Effect meets Shadowrun) or gritty post-apocalyptic. The system works pretty well, with some 4e influences in the way of "encounter powers" (powered by tokens, a la Savage Worlds). The current problems are the damage system (it's really easy to die) and making the numbers work in what is essentially a d% game. The big plus for the game is that there are no XP, and characters do 95% of their improving through actual play - the equivalent of class abilities and feats are purchased only rarely, as PCs level up their skills to required minimums.
I think, if you want to make your own rules set, you need to know what you want to get out of it. If you're just making a system that "makes sense", it'll fall apart or be hugely derivative (and if you're spending time making a derivative... why not just play the original game and save yourself some time?). I'd just write down on a piece of paper your main goal. For the three big ones I've made, the goals were:
[sblock=design goals for my games]
1) Mutant Game:
a) Character Generation based off random rolls for powers/drawbacks
b) Fairly open character classes that could be viewed multiple ways
c) Characters that are killed more easily than in standard d20 D&D (a heavy reliance on massive damage was used).
2) d12 System
a) Really fast skill resolution (roll the required number of d12s, take the best two results, doubles meant a critical success or failure, depending on the number rolled)
3) d% system
a) Remove the need for levelling your character and slowing down the game by increasing your level.
b) Keep the math as simple as possible (this was actually pretty easy to do, even with percentile modifiers!)
c) Every player should only bring two dice to the table.
d) Make sure that each PC was different and specialized in different zones, despite being a classless system (there are around 40 skills, and about half of those are useful in every session... "necessary" skills like Dodge and Perception are natural skills that every PC has, though every player must choose where to focus training in their natural skills).[/sblock]
Also, there are a lot of people who have made their own RPGs, and some are downright FUN. I'd recommend any RPGer check out this site:
http://www.darkshire.net/jhkim/rpg/freerpgs/index.html
It's a list of free RPGs, many made by random joes trying something out. Some are good, some suck, but if you want to make your own system, it's a good place to look for inspiration.