In Plotkin's 'The $\lambda$-Calculus is $\omega$-Incomplete' (The Journal of Symbolic Logic Vol. 39, No. 2 (Jun., 1974), pp. 313-317), an example is given of two (untyped) $\lambda$-terms $M$ and $N$ such that for each closed $\lambda$-term $c$, $Mc$ and $Nc$ are $\beta \eta$-equivalent (i.e., convertible), but $M$ and $N$ are not $\beta \eta$-equivalent.
The example he gives is very complex. What is the simplest (or a somewhat simple) example of such $M$, $N$?