The stress distribution you highlight is not universal - different design codes allow different assumptions about the stress-strain characteristic.
For example, in the UK, BS 5400-4, which was until relatively recently the design code for concrete bridges, specified a stress-strain characteristic looking like the following. Note that this shows the concrete reaching a strain of 0.0035, and includes a parabolic portion at low strains.
In certain circumstances (rectangular beams, or where the portion of the section that is in compression is rectangular) BS5400-4 allowed a simpler assumption: “the compressive stress may be taken as equal to 0.4fcu over the whole compression zone”.
BS8110-1, which was the design code for reinforced concrete in buildings, had the same parabolic stress assumption, but a different simplified rectangular assumption:
The current design code (BSEN1992-1-1 for buildings and BSEN1992-2 for bridges though the latter refers to the former in this respect) (aka Eurocode 2) has other assumptions (with various equations to calculate the critical values on the graphs):
For simple cases, the different assumptions give substantially the same answer (if the code authors have done their job right). For the Eurocode cases it’s possible to examine this algebraically relatively simply, because all the equations are using a compatible approach. For the case of a rectangular section and C40/50 concrete designed to BSEN1992, using the parabolic assumption provides a concrete force of 17/21 x b f.cd (where x is neutral axis depth, b is section width and f.cd. is the design value of limiting concrete strength) and the rectangular assumption provides (with less arithmetic) 0.8 x b f.cd, which is about 1% lower. However, it’s not quite that simple because the point of action of these forces are in different places (slightly), and if you follow that through you find the rectangular assumption (for these cases) always gives a slightly greater bending strength for the section, the magnitude of the difference varying depending how much of the cross-section is in compression, but the difference is no more than 0.6% over the allowable range of parameters. I've done this previously:
All these examples agree with respect to not counting any concrete in tension, and (as noted in NMech’s answer) this is because the behaviour of concrete in tension is even more unpredictable. However, these examples are all used for strength calculations. Some codes (including these) do include the stiffening effect of concrete in tension when calculating stiffness or crack widths – so that implies the existence of more stress-strain assumptions within these codes.
Further, many engineers actually take the path of (conservatively) disregarding that stiffening effect even when it is permitted, which introduces yet more de facto characteristics (not formally codified). (It should be noted that reinforced concrete sections are rarely critical in stiffness, so these assumptions normally do not actually affect the outcome of the design – if they did, those engineers should be taking more care).
To answer the question directly: the various different assumptions are all intended to provide a basis for design that is pragmatic and safe. They try and reflect the true behaviour of a highly variable material (the codes need to work for a wide range of constituent materials and strengths) as closely as possible without being unworkably complex to operate or inefficient.
0.85 is just what whichever code that graph has been taken from uses to get a suitable and safe set of assumptions.