An SST-90 with awesome heatsink could be overdriven to give 3000 lm. A good aspheric lens of >40mm diameter could easily produce the 6 degree (half angle) beam. 54W is plenty of power to produce this output. I don't think this is at all outside the realm of possibility.
On the other hand, there are some fishy claims, but they point to poor marketing not poor design. The 3000/2000/1000 mode changes are stupid. The difference between 3000 and 2000 lm is barely noticable. 2000 to 1000 is noticable, but not all that big. This suggests that the marketers and/or designers haven't ever used the lights, or anything similar, with any idea what the specs mean. Having built and dived with my own lights, I can say that 3000/1000/300 would be a much more useful range.
Also, they claim 3000 lm for 2 hrs, and 2000 lm for 4 hrs, then 1000 lm for 6 hrs. This doesn't make any sense from an engineering perspective. In an ideal world, the 2000 lm burn time should be 3 hrs, not 4 hrs. But realisticly, 2000 lm should give a little more than 3 hrs, and 1000 lm should give more than 6 hrs.
To summarize - the claimed specs don't seem to be outrageous, but seem to be driven more by marketing than by engeneering.