Collecting some of the answers above, and editing a bit:

> What can one say about its torsion elements?

$\newcommand{\Mod}{\mathrm{Mod}}$As with lattices in Lie groups, and in word-hyperbolic groups, the mapping class group of a surface has finitely many torsion subgroups, up to conjugacy.  This means, in principle, if you fix $g$ and $b$ then you can list all finite subgroups of $\Mod(S_{g,b})$.  This follows (for example) from "Nielsen realization" stating that for any finite subgroup $G$ in $\Mod(S_{g,b})$ there is a hyperbolic metric on $S$ where $G$ acts via isometries. 

Despite this one should note that for _any_ finite group $H$ there is a $g > 0$ so that $H$ is a subgroup of $\Mod(S_g)$.  

Also, note that if you require that the surface has boundary, and require all mapping classes and isotopies to fix the boundary pointwise, then the finite order elements go away.  For example, the braid group has no torision. 

> Is it true that every mapping class of order 2 is orientation reversing?

$\newcommand{\ZZ}{\mathbb{Z}}$No. In the closed case the most famous examples are the "hyperelliptic involutions".  As a somewhat different source of examples, take any surface $S$ and take any double cover $T$.  Then the deck group of $T$ over $S$ is a copy of $\ZZ/2\ZZ$ inside of $\Mod(T)$. 

> Now assume that $S$ also has marked points on the boundary, and homeomorphisms are also allowed to permute them. Is it true that every mapping class of order 2 is orientation reversing?

Again no, as above.