The answer is $\frac1{12}$, independent of the shape of triangle.
To see the independence, let $T_0$ be a fixed triangle with unit area.
For any triangle $T$ with unit area, let $\eta : \mathbb{R}^2 \to \mathbb{R}^2$ be an affine transformation which sends $T$ to $T_0$.
Recall affine transformation preserves ratio of area of geometric shapes. If $p_1, p_2, p_3$ are $3$ points in $T$, then $\eta(p_1),\eta(p_2),\eta(p_3)$ are three points in $T_0 = \eta(T)$ with
$$\verb/Area/(\triangle_{p_1p_2p_3}) = \verb/Area/(\triangle_{\eta(p_1)\eta(p_2)\eta(p_3)})$$
Furthermore, if $p_k$ are sampled uniformly from $T$, then $\eta(p_k)$ will be uniformly distributed over $T_0$. This implies the expected area of
"uniformly sampled" triangles in $T$ equals to that of $T_0$.
Since this is true for all $T$, the answer is independent of shape of triangle $T$.
This problem of finding the expected area of "uniformly sampled" triangles in a triangle is known as Triangle triangle picking, see the entry on mathworld for more information.
There are various generalizations of this problem. Instead of $3$ points, one can uniformly sampled $n$ points from a triangle of unit area and
askes: What is the expected area of the convex hull formed by those $n$ points?
This is part of the Triangle Point Picking problem, see entry on mathworld again for more information.
The answer of the expected area of convex hull of $n$ points is given by the formula:
$$\Delta_n = 1 - \frac{2}{n+1}\sum_{k=1}^n \frac1k = 1 - \frac{2H_n}{n+1}$$
where $H_n$ is a harmonic number. It is possible to write down a double integral to compute this number. For a derivation, see the second half of my answer to a semi-related question.
In any event, for the problem at hand, the expected area of "uniformly sampled" triangle is
$$\Delta_3 = 1 - \frac{2 H_3}{3 + 1} = 1 - \frac12\left(1 + \frac12 + \frac13\right) = \frac{1}{12}$$