Don't use electrolytic caps - I'd include tants there - because avoiding leakage currents, really, really matters/will corrupt the bit current definition. And these are current-output dacs.
Also note that these are ripple filters for the internal (bit) current sources running at your OS rate, so need to work at min. 176Khz (and above) in a 4xOS unit like Naim's (and other implementations with the SAA7220B) - therefore much larger C values won't help, neither does large physical size (more ESL). Naim's tight-layout use of stacked-film over a fat ground-plane is spot-on when you think about it in those terms.
Increase the value on the MSB on each channel - all the others are of course divided-down from this, so you only have to change two, to affect the lot. Up to 1uF film providing in a very compact package is, iirc, what I went for. Something like that anyway, I suggest.
PS for all that, in my Naim CD2, I would not call the alteration remotely audible AFAICR. IN something with less-well resolved layout and parts choice -who knows. Given a free hand, I'd be looking at SMT parts on a layout like Naim's for the ground plane for sure, for best HF performance.