The AD9958/9959 evaluation board suggests separate AVDDs for the individual outputs, connected via zero-ohm bridges. This concept is completely absent in the AD9958/9959 datasheets where all those pins are simply marked AVDD. In my AD9958-based design, I buffer AVDD with 10nF/100nF pairs in the four locations that are closest to the analog outputs in use (please see attached image; capacitors marked with yellow X). The blue copper plane in the image is AVDD (it's a 4-layer design). Am I right in assuming that I can skip additional buffering on those AVDD pads that are associated with the two missing channels (marked Y)? It would save me a few parts on that side of the chip, and hence also precious space on the PCB.
As to my application, quality of the output signal is of virtually no importance. Both channels run at the same frequency (no risk of crosstalk), and the next thing that happens after lowpass filtering is conversion to a rectangle signal via a LVDS receiver chip.