Hello again, Adrian,
I think KJDS was "pulling your leg" when he asked about the 0.1% resistors! He was really making the same point - really, you just don't have to by that precise in all this. Sure, the constant current circuit that you have will work, and no doubt beautifully, but it's overkill, that's all.
When manufacturers give ratings to their products, they are usually pretty conservative, and figures are specified over a wide range of temperatures and typical component tolerances. There is a good safety margin built in. An LED rated at 22mA will happliy run at 25, even 30mA for many thousands of hours. If you are experiencing LED failures, I will bet my best shirt on the fact that it is not triggered by minor current overload against some specification figure.
Even different LEDs from the same manufacturing batch will perform a few % differently with respect to one another. Put a few in series and parallel, and you will find minor variation in the voltage across each LED, across each string, and also differences in the current flowing in each string, that make trying to drive an exact current through each LED nigh on impossible. But they'll all still work fine.
Now, such a small current LED, accidently exposed to, say, 1A even for just a few mS may well fry. That will be enough energy and time to burn the delicate LED die out. A simple wiring fault or incorrect component in a driver circuit could easily lead to that.
Here's another thought. Our ears, eyes, and for all I know, probably our noses respond logarithmically to changes in sound/light and smell. This is rather handy, as it enables them to operate over a HUGE RANGE of values. Compared to the very quietest sound you can hear and the point at which the sound is so loud it actually hurts, the ratio of sound pressure is 1:1,000,000,000,000 - that's a million million! And this is one reason why we measure sound levels in decibels, instead (a much more manageable measure). Moreover, ears can bearly detect a 1dB change in sound level, and yet this corresponds to a level change of 25% in the absolute level of the signal.
Now, I don't know what the threshold is for eyes, but I will put another of my shirts on the fact that, if you built a circuit to drive your LED with first 20mA for 1 second, then 22mA for a second, and then repeat this, your eyes would not be able to see the change in level of illumination. And even if they could, it would be so small that, when shown either brightness in isolation, you'd hve no reall idea whether you were looking at 20 to 22mA.
Bottom line is - yes, you can specify and build to this level of precision, but it yields no advantages in the workings of your circuit, whatsoever. And, at the risk of running out of shirts alltogether, I'll state that no 20mA-rated LED will give up the ghost at 23mA. There, I've said it.
Please don't let me put you off - Keep experimenting and Learning! It's all good.