I'm a proponent of addressing "should" after "could". You could go the other way (check if any of the arguments is a label, as defined by r@key) fairly easily:
\documentclass{article}
\usepackage{natbib}
\bibliographystyle{whateverstyle}
\usepackage{etoolbox}
\usepackage{amsmath}
\usepackage{hyperref}
\usepackage{cleveref}
\usepackage{autonum}
\makeatletter
\def\omniref{\@ifnextchar[{\citet}{\@ifstar{\citet}{\omni@ref}}}
\def\omni@ref#1{%
\newif\iflbl\lblfalse
\def\do##1{\@ifundefined{r@##1}{}{\lbltrue}}%
\docsvlist{#1}%
\iflbl\cref{#1}\else\citet{#1}\fi}
\AtBeginDocument{\@ifpackageloaded{autonum}{\autonum@generatePatchedReferenceCSL{omniref}}{}}
\makeatother
%
\begin{document}
\bibliography{refcite}
\begin{equation}\label{aeqn}
y=x
\end{equation}
\omniref{vanLoosdrecht99} found that \omniref{aeqn,eqn2} can hold.
\begin{equation}\label{eqn2}
y=x
\end{equation}
\begin{thebibliography}{1}
\bibitem[{van Loosdrecht and Henze(1999)}]{vanLoosdrecht99}
van Loosdrecht, M. C.~M., Henze, M., 1999. Maintenance, endogeneous
respiration, lysis, decay and predation. Water Science and Technology 39~(1),
107--117.
\end{thebibliography}
\end{document}
Edit
Assuming the bibliography entries are defined as b@citekey, the approach you proposed is doable as well. If all of the entries are cited in the bibliography, the following will use \citet; if not \cref will be used:
\documentclass{article}
\usepackage{natbib}
\bibliographystyle{whateverstyle}
\usepackage{etoolbox}
\usepackage{amsmath}
\usepackage{hyperref}
\usepackage{cleveref}
\usepackage{autonum}
\makeatletter
\def\omniref{\@ifnextchar[{\citet}{\@ifstar{\citet}{\omni@ref}}}
\def\omni@ref#1{%
\newif\ifbib\bibtrue
\def\do##1{\@ifundefined{b@##1}{\bibfalse}{}}%
\docsvlist{#1}%
\ifbib\citet{#1}\else\cref{#1}\fi}
\AtBeginDocument{\@ifpackageloaded{autonum}{\autonum@generatePatchedReferenceCSL{omniref}}{}}
\makeatother
%
\begin{document}
\bibliography{refcite}
\begin{equation}\label{aeqn}
y=x
\end{equation}
\omniref{vanLoosdrecht99} found that \omniref{aeqn,eqn2} can hold.
\begin{equation}\label{eqn2}
y=x
\end{equation}
\begin{thebibliography}{1}
\bibitem[{van Loosdrecht and Henze(1999)}]{vanLoosdrecht99}
van Loosdrecht, M. C.~M., Henze, M., 1999. Maintenance, endogeneous
respiration, lysis, decay and predation. Water Science and Technology 39~(1),
107--117.
\end{thebibliography}
\end{document}
\end{document}
Yields:

As for the "should": Obviously this deviates from convention and may make interpreting errors more difficult. The biggest potential issue is if you have a label and a citation with the same key, which may result in unexpected behavior. However, if you are using sufficiently different naming conventions for labels and citations such that you would still be able to infer the intention of the command and eliminate overlap (e.g., eqn:somthing, fig:somethingelse and Doe98) and you are aware of the other potential issues, I'd say go for it.
Edit
Since \citet accepts options and has a starred variant but \cref does not, an additional line was added to each to differentiate the two (if \omniref is followed by a star or has options, the \citet is used, otherwise the keys are compared). Also, missing %s were added to remove the additional white space.
Edit
As per the comments below, added compatibility with hyperref and autonum.