I agree with Steve K that stability can be a problem, and that a resistor per string is probably better than the current mirror. You need to plan on half a volt or more across the resistor for it to work properly. But you may find that with a proper driver and construction, you don't even need the resistors.
I'm quite surprised to see that the Vf of the LEDs is 3 volts or more. Greens are usually closer to 2 volts. Do you have a part number for us to look at? The rest of this post will assume that the 3.0-3.4V spec is correct. If it isn't, then ignore most of this and I'll have some other comments.
I notice that the Vbe of the 2n5088 is more like 0.7-0.8V at the design current (around 70 mA). This is going to drive a lot more heat into R3, which should probably be a 5W resistor, not 3W. It should also be 0.75V/4.5A = 0.166 ohms. The nearest readily available values are 0.15 and 0.18 ohms.
Vbe of the 2N6715 is 1.2V at 1A. It will be pretty near this at the design current of 0.75A. Add this to the drops across R3, Q2, and R4 (unspecified) and you get over 2V. You see that you are rapidly running out of voltage to run the LEDs with a 12V supply.
Even worse, since this is to be run off a deep-cycle battery, you want it to run down to 10.5V or so. This is barely enough to run the LEDs, even without considering the losses in the regulation circuitry. I think you really need a low-dropout regulator. The schematic for one of my proven low-dropout regulators can be found here:
http://www.candlepowerforums.com/vb...10A-linear-LED-driver-New-and-Improved!/page2
(Check post #33)
This can be simplified somewhat in your case, as you only need one 'mode', and probably don't care about precise regulation as battery voltage changes (plus or minus 10% is normally not detectable to the human eye). I can help you with simplifying it and selecting appropriate parts if you want to build one. Total voltage lost across this circuit could easily be 0.3V or less, which is perfect for your application.