If you can have enough room above plants, you can basically use any reflectors with a narrower beam pattern (i.e you can move the light far away to make sure the plants don't get too much light). You could try something like 60-90 degree beam angle, but you can add the reflector later, too. 90 degree means 45-degree in both side, so from 2 feet distance, the circle diameter is around 2 feet.
I would say that try CXB3070. You'll gain a bit more output from CXB3590 (for the same watt), but I'm not sure if it justifies the price difference (unless you are a efficiency-number freak like me).
CXB3070 4000K BD bin (1.4A, 50W, 50C Tj) gives 7989 lumen according to:
Cree Product Characterization Tool
Since it is directional (unlike T5HO), you'll get more than 2x light over your current T5HO setting, so it should cover 2'x2' area. One glitch is that the coverage area is not as even as T5 (high light area in the center).
From my calculation, lowest CRI and higher color temp gives more PAR in general. But in terms of yield photon flux (YPF; i.e. after putting weight representing that red light is more efficient than blue light for photosynthesis (PS)), the lower color temp isn't so far behind (i.e. color temp doesn't matter too much). Higher bin is always better if you can find them.
Here are the relevant values if you want to calculate by yourself.
[Edit: oops, I posted this info in p.2 of this thread. I have forgotten about it, so this is just a subset of the previous table.]
## ppf.per.lum ppfd.per.fc ypf.per.lum ypfd.per.fc
## Cree.3000K.80CRI 0.01424752 0.1533591 0.01288196 0.1386603
## Cree.3000K.90CRI 0.01407376 0.1514886 0.01230506 0.1324506
## Cree.4000K.80CRI 0.01407376 0.1514886 0.01230506 0.1324506
## Cree.5000K.70CRI 0.01382456 0.1488064 0.01204868 0.1296909
Basically, you can get the lumen (or lumen per watt) for each CRI/K model from the Cree datasheet. Note that higher K generally have higher lumen values (and higher bin numbers). Then you can convert the lumen value to PAR photosynthetic photon flux (PPF; use the multiplier in par.ppf.um) or to YPF (use ypf.per.lum column). The value calculated has the unit of micromol/s. The other 2 columns can be used to convert the foot-candle reading from light meter to PPFD or YPFD (unit will be micromol/m^2/s). PPF"D" and YPF"D" represent "D"ensity. So PPF and YPF are the total output from the fixture, PPFD and YPFD are the density of these light on the leaf (i.e PPF per m^2). This is similar to lumen (total output) vs lux/fc (density).
Now, one thing to note is that PS is not everything. I'm not completely familiar with the physiology of CP, but red light is likely to be important for phytochrome related phenomena. I'm guessing that day length influences the seed and plant dormancy of Saracenia. So red and far red ratio becomes relevant. I have been using 3000, 4000, and 5000K, and all seems to work well for orchids. I subjectively (i.e. without strong scientific support) choose that 4000K is the default for me (it just seems to be a good balance).
Something like this:
Cutter Electronics, Supplier of lowest cost leds on the internet
I think that their coupon is still effective (I think I posted earlier in this thread), but you need to see if it is cheaper domestically once you consider the shipping cost. cutter seems to have top bin frequently, which is difficult to get domestically.
With regard to the driving current, you'll probably need at least 1400mA (50W) if you want to get that much of light from 1x COB. Maybe LPC-60-1400 is a choice: LPC-60 datasheet
If you want to get more light (sacrifice a bit of efficiency) you can go with LPC-100-1750: LPC-100 datasheet
It gives 1750mA (63W), and you'll get 9599 lumen.
These drivers are not dimmable nor highest efficiency, but they are cheap (and good quality).
If you want to go dimmable (1.17-1.95A = 40-70W), HLG-80H-42 is one of the best: HLG-80H datasheet
I personally would go with the cheaper LPC for a single driver + single COB, though. If you are going to connect multiple COB LEDs to a single driver then HLG could be a good choice.