for the sake of simplicity, let's suppose your leds run 1A. 3W LED like that are resonably common.
You are getting 12V from your power supply, with resistor it is a series circuit everything has the same current flux going trough it, you are pulling 1Amp@12V form the psu. That is 12W in teletubbies world.
In real world that means pulling 12W/efficiency from your mains. A (very) good PSU might reach 85-90% overall at its sweet spot on all rails. Pulling just an amp from a single 12V rail, it will probably mean something in beetween 35% to 65% efficiency as the rail is both massively under the expected load and the load completely unbalanced (your 5, 3.3 and al the other one are unused). You will possibly get less that that on dodgy PSU, I'd be very surprised by more on anything.
In real world, at low drain, psu's tend to be massively over the target voltage, specially cheap ones. I have seen 12V rails at 13.5, it is not good, but it is usually not the end of the world, specially as under load the voltage sags to something more acceptable. So in reality you have 13W, that cost you beetween 20 and 37W to produce, and all that goes to actual light beeing produced using only using 9W. IT is not going to be 9 watts of light as the LED are going to heat a fair amount, and produce useless far infrared (unless you plan to run your LED at 900C)
Your LED are probably around 3V fw, with a range in beetween 2.8 an 3.4 or something in this ballpark. so you can only fit 3 in series if you want full power, giving you a range of 8.4 to 10.2 V . You need to dissipate the rest somehow, trough a resistor. Basically it is a badly inneficient PSU to power a not very efficient solution.
Any linear regulation will be inneficient for driving LED, unless your LED forward voltage can be guranteed to be very close to the PSU voltage.
As you cannot guarantee PSU output voltage and LED forward voltage spread is relatively high, you need to have quite a lot of safety margin, which equates quite a lot of dissipated energy.
A ballast that is designed just for the purpose of driving a 54W tube is likely to be more than 90% efficient. The tube itself is less good than led at producing useful energy for the plants, but with your 635nm led you are quite off anyway. That is how I reached the conclusion.
In my case I was planning on strings of 120 low power led with a linear regultor (mosfet + adjustable voltage reference) driven straight from rectified and filtered mains. The cost of putting together this system makes it usefulness dubious at best.
Hope I helped