The risk with that kind of tech tree is that it can easily be boring.I don't mind that at some point you only get into infinite standard researches but before that there should be some spiced up techs that stand out (maybe with some rule that you at least get one of a certain bunch of techs to keep it fair).
I always assumed this would be the way it would be... I mean, before you can upgrade "Generic Swords Upgrade I" you'd have to research swords to begin with. It's just that once you've actually researched the first stage, you can continue researching that tech for improved results.
Logarithms can be pretty harsh. Instead of "Benefit = log(level)", sometimes it works better to have "Benefit = root(level)", so that you don't end up in a situation where diminishing returns diminish too fast. The base of the root is arbitrary.
This is my preferred route, too - at least over logarithms. The diminishing returns of logarithms (and asymptotic approaches) is that you reach a point where there is just no reason to keep going. Even if you have resources flowing out of your ears and more than enough to throw away, you'll never notice that +0.01% boost that you paid 100,000 gold for. It just becomes a waste of space and they may as well have implemented a hard limit.
Personally, above all these methods I prefer a method without diminishing returns. I would prefer if progressive generic bonuses provide constant bonuses but with increased cost. Whether the cost should increase as a root, linearly, or exponentially I would leave up to testing. My reason for this is that I think this method allows for research to always remain a factor in the game - with diminishing returns there comes a point on larger maps that research simply stops being relevant.
Another downside to diminishing returns or costs rising too quickly is that it encourages you, for the sake of efficiency, not to focus too much on any one tech. And the result of that is, after enough time everyone ends with more or less the same research other than faction-specific or randomized techs. Allowing people to continue researching fields that are particularly important to them without too much of a hit to efficiency allows for much more variability in technology research.
I find too many games give increasing returns for research, instead of decreasing - for example, a tier 1 farm produces 100 food, a tier 2 farm produces 300, and tier 3 produces 900. That's not really balanced or realistic, but I expect this game to be more subtle.
I don't see why it's unrealistic. Look where we were 300 years ago, and look where we are today. Technology has absolutely exploded - we have had increasing returns. Computing is a perfect example, even though that is slowing down. Still, once the next major step into a whole new method of computing is made, it'll likely happen all over again. Not to mention what's realistic isn't so important to me in terms of game design.
And even in the case that such a method often results in 'unbalanced' situations, I think that it could be made to be balanced enough to satisfy anyone but the diehard multiplayer people who are hell-bent on ruining every game they get their dirty little hands on with their obsession with absolute balance