So.... if the magenta trace is 25MHz with 9nS rise time, it makes sense that 30MHz at 10nS will be rounder than that.
Basically yes. At least at/near the limit you rather need to compare frequencies in ratio to the sampling rate. I.e. you rather need to compare the 24 MHz @300MSa/s from the SDG to 20 MHz (not 30) @250MSa/s from the HDG.
The SDG has a clear advantage, though, in this regard. It has the higher sampling rate.
And besides that, its 4x interpolation/oversampling capability certainly relaxes the requirements for the analog reconstruction filter.
(Don't know if the oversampling is used for anything else - the docs don't seem to reveal that.)