Ok...slowly i understand the circuit...and your calculation.
Due to evp's advice i will have to switch the led!!!
Just because of the price - the shop sells it for 2,14€ vs. 10€ this saves me so much money! - and this comes in handy with our problem (when i got everything right).
If i switch to this one http://www.satisled.com/3w-rgb-led-emitter-on-star-1w-for-each-color_p234.html i'll deal with 3.6 max v loss. So just to be sure i understood this correctly i have 5 - 3.6 - 0.65= 0.75 V Vsat. To get the red one up to the same V i burn 1V with a 2.8 Ohm resistor. The friction in the transistor is reduced to 262mW.
So lets give me another try with a part 
http://www.farnell.com/datasheets/1504602.pdf
Vsat is -500mV (i can provide 750 in the worst case), Powerdissipation can be over 1W (i am below with 262mW) Did i forget something?
Just to ask it: is it really that bad to drive an led via resistor without ccs??? I can't imagine how http://www.satisled.com/10w-9w-rgb-led-emitter-3w-for-each-color_p236.html those are driven... (they get special drivers i guess as dc42 mentioned in his first post).