"Game-ready topology": how many triangles should we aim for?

Back in the day, when I were a lad,

actually, when I was a bit older, when I was a lad ACSII art was how we created graphics. 

The number of tris, was a major consideration in the earlier days of modern 3d graphics,

but with the technology today, it is not so much of a consideration to aim for, as low as possible number of tris.

The acceptable number of tris in my opinion is as low as you are willing to go, without losing the desired shapes and detail.

I would say however, if you are developing a game, and wish to distribute it to as many game players as possible,

Consider how many assets are in the scene,

if your game is a tank battle game, featuring only 2 tanks, and not a lot else,

then think an acceptable amount as mentioned.

Then you will also need to think about who will be playing your game, and the hardware of those players.

If you develop a game on your top tech hardware machine, and it works well, it might not work on a machine with less capable hardware.

This is why today, the number of tris is considered. Games are usually designed and built using powerful hardware, then tested on much less powerful hardware. Developing games for Xbox or Playstation is much easier as the hardware for each player is basically the same.

Game development companies will consider, and decide on a target number of tris for each model in the 'game design document', which is produced during pre-production.

So, in conclusion, 

There is much more to consider than one asset at a time as you are modeling each one, and pre-production is vital to producing a successful game.

However, if its just for you and a couple of mates, then you can play around and make any changes on the fly.

Happy Blending!

Adrian's answer is on-point, in that you should think of the polycounts (and everything really) in terms of the desired outcome and final project as a whole. Also I'd add that you want to consider the role of the object in-game and the scale/distance it will be used at; if the object will always be in the far distance, then there's no reason to model, f/e, a high-poly moon in the sky when a simple card/plane will do for the purpose. Your main characters and enemies will have many more details and resolution than a cup on a desk, which is where the design doc comes in, as stated in Adrian's post.

In the words of Einstein: "It can scarcely be denied that the supreme goal of all theory is to make the irreducible basic elements as simple and as few as possible without having to surrender the adequate representation of a single datum of experience. "

Or, as is often paraphrased: "Make it as simple as possible and no simpler." I'd go one step past that and say "Make it as simple as NECESSARY and no simpler." You can spend a lot of time optimizing just to do it. Are you going to spend a ton of time ensuring perfect quad retopo work on a background object that won't even be animated/destructed? You COULD do it, but probably shouldn't because that time is better spent elsewhere.