They use FPGA's for possibly two reasons:
1) because it's maybe the only way to get the high performance for the overall system cost they are targeting.
Even the fastest processors are no good for modern real-time DSO sampling architectures, so that only leaves an FPGA, or to spin your own ASIC.
Yes, FPGA'S are expensive devices, but the NRE cost is zero.
Ballpark calcs:
I don't know how many units they be targeting, but lets say it's 100K units.
An ASIC might cost you say $1M. So that's $10 per unit just for the NRE alone, not including the actual chip.
The Xilinx XC6SLX25 used in the DS2000 for example is $55 one-off. Not sure what it is for 100K units, but lets say $10. So similar price to the ASIC when you include the chip cost, testing, etc. Then you can't use the same part again in the design for something else (The DS2000 has 3 FPGA's)
And then there is the benefit of being able to update the thing in-system, if you find a bug, or some competitor comes out with a new feature that you want to match. That alone is enough reason to pay more for an FPGA (if it is actually more expensive).
Dave.