Apparently, at least with Rigol gear, heat sinking can vary widely. I was lucky with my MSO4000, manufactured early 2016, it has a substantial single-piece heat sink installed on the acquisition engine (covering four MXT2001 ADCs and four Virtex FPGAs). The fifth Virtex has an individual heat sink bonded on as well as the MSO piggyback FPGA (Spartan??):
Other owners of DS/MSO4000 series instruments reported several different cooling approaches on their instruments from almost non-existant to quite decent (individual heat sinks on all the FPGAs but not on the ADCs) Of course, if my unit, which AFAIK is the only reported one with the big, single heat sink, has installed proper thermal pads on all the semiconductors to transfer the heat, is still unknown. I didn't bother to rip the heat sink off...
Coherent with this is the cooling situation of my DS(MSO)2000A, individual heat sinks on all the FPGAs but not on the MXT2001 ADC. Considering that the ADC dissipates a maximum of 2W and is attached to the PCB with an additional thermal pad (25mm²), one may get along with this if an adequate PCB thermal design is applied, especially since this instrument has a single such ADC and and not three more fending in the neighbourhood (like the DS4000). Yet, the FPGA on the signal generator piggyback board is left without a heatsink. Maybe it's a small(er) unit that doesn't dissipate that much heat.
Unfortunately, I cannot report on Siglent or other, more recent scopes. But at least the Spartan 6 in the Siglent SSA3000X spectrum analyzer is running without additional heatsink. Presumably, the workload on the FPGA in this application is less than on the components of the acquisition engine of DSOs.