Too long for a comment. Actually I do not think that it is written anywhere but these kind of counterexamples are usually provided by the Ornstein-Uhlenbeck semigroup, generated by $\Delta+Bx \cdot \nabla$, where $B$ is a matrix. Assuming that all eigenvalues of $B$ have negative real parts, then an invariant measure $\mu$ exists (and is given by a Gaussian density). It turns out that the angle of analiticity in $L^p$ of the invariant measure can be computed exactly and can be smaller than $\pi/2$, even for $p=2$. This can be found in a paper by Chill, Fasangova, Pallara and myself. 

<cite authors="Chill, R.; Fašangová, E.; Metafune, G.; Pallara, D.">_Chill, R.; Fašangová, E.; Metafune, G.; Pallara, D._, [**The sector of analyticity of the Ornstein-Uhlenbeck semigroup on \(L^p\) spaces with respect to invariant measure**](http://dx.doi.org/10.1112/S0024610705006344), J. Lond. Math. Soc., II. Ser. 71, No. 3, 703-722 (2005). [ZBL1123.35030](https://zbmath.org/?q=an:1123.35030).</cite>

To obtain similar examples in unweighted spaces one has only (but patiently) to compensate the weight thus obtaining an operator with a linear drift and a quadratic potential. This works however only from dimension $2$ on; in the one dimensional case the operator is always self-adjont but still the angle of analitycity is different from $\pi/2$ in $L^p$, for $p$ different from $2$. There is also a paper by E. Priola and myself dealing with non-analytic Markov semigroups where one finds other examples. 

<cite authors="Metafune, G.; Priola, E.">_Metafune, G.; Priola, E._, [**Some classes of non-analytic Markov semigroups**](http://dx.doi.org/10.1016/j.jmaa.2004.02.037), J. Math. Anal. Appl. 294, No. 2, 596-613 (2004). [ZBL1067.47055](https://zbmath.org/?q=an:1067.47055).</cite>

It would be nice to have a direct approach, avoiding the detour.