Key references and conrefs work in both DITA-OT and Native PDF Publishing. I haven't tried conkeyrefs yet. Variables (keydefs) work fine in both DITA-OT and Native PDF.
DITA-OT has always supported use-by-reference footnotes, but I understand that v4.4 of AEM Guides has fixed that oversight in Native PDF Publishing. I cannot confirm since we haven't upgraded to 4.4 yet.
DITA-OT does require more technical/code expertise. Native PDF is MUCH easier to create output templates for and there are several things I've been able to do in Native PDF Publishing that I could not in DITA-OT (outputclass to wrap text around selected images; outputclass to reduce the font size selected text). It was also much easier to design and implement front and back covers in Native PDF Publishing.
I have not yet figured out a way to suppress table frames/lines on command for select tables in Native PDF Publishing, and you could do that in DITA-OT.
Numbering heading levels in DITA-OT required a lot of programming. It was "relatively" easier in Native PDF Publishing, but that also took some time and experimentation. It just took a lot LESS time in Native PDF Publishing.
Having used DITA-OT, FrameMaker Publishing Server, and Native PDF Publishing, Native PDF Publishing wins hands-down. It is so much easier.
But, if your DITA-OT output templates are already designed, there's no reason not to implement those as well.