The bias current has to be supplied by the whatever circuit is connected to the opamp input. The source impedance of that circuit then determines the voltage.
It is irrelevant if the nominal voltage of that node is 0 V.
Specifically for your example:
Looking out of the non-inverting input of OA3, we see, in parallel, the output of OA1, and R8 connected to OA1's inverting input. The output of OA1 is low impedance, especially so because of feedback. It is reasonable to assume that this node is close to 0 Ohms. Hence the \$I_{b+}\$ bias current of OA3 does not play a role. It is simply provided by OA1, without any other effects.
At the inverting input, things are slightly more complicated. Again, the impedance we see is R4, in parallel with R5. One might be tempted to conclude that the impedance thus is simply R4 || R5 = 9 kOhm. However, because of feedback, the inverting node must remain at the same voltage, so the extra current must go through R5 exclusively. It will create a voltage error of \$I_{b^-}\cdot R_5\$. Another way to see this is that for the \$I_{b^-}\$ bias current, OA3 is wired as a transimpedance amplifier.
See this for more info: http://www.ecircuitcenter.com/Circuits/op_ibias/op_ibias.htm