Both capacitors are noise filters. C75 works with the Thevenin equivalent of R124 and R178 to form a single-pole lowpass filter. You want this corner frequency to be significantly below the lowest signal frequency of interest.
C40 both filters noise on the power supply rail, and lowers its impedance caused by wiring and pc board trace resistances and inductances. Typically you want one such large capacitance per xx square inches of circuit area, plus smaller capacitors right at the power pins of each IC. The standard part is a 0.1 uF ceramic cap with the shortest possible leads/traces to the chip pins, one cap per power pin.
Check your numbers - 50 mA is way too high for the linked schematic. Also, most of the power supply current does not go into/out of the virtual GND. The GND circuit output sees only the algebraic sum of all circuit ground currents.
Update:
C40 filters noise on the power rail, but more importantly it filters noise caused by the signal going through the opamp. A portion of the current at the power pin is the static current needed to operate the internal circuits, but another portion is the amplified signal current going into and out of the output pin. If the supply rail impedance at the IC pin at the signal frequencies is not zero, the signal current will appear as a varying voltage at the power pin. This varying voltage can now appear across the opamp's input circuit as an unwanted form of feedback, sometimes causing the circuit to break into oscillation. This is especially true with linear audio power amplifier chips. The LM386 makes a fair AM radio transmitter.
The non-zero impedance of the power connection comes from both the output impedance of the source (at signal frequencies, not just at DC), and wire and pc board resistance and inductance. Because electrolytic capacitors usually do not have good high-frequency performance, a compromise is to have large caps in the area for low frequency decoupling, and smaller caps with better high-frequency behavior right at the IC pins to get the impedance at the signal frequencies down as low as possible. These caps also supply some signal current during transients, so the size does matter. 0.1 uF is the standard value for general-purpose work at low signal currents. At 10 mA and above I jump to a 1 uF ceramic at the pins. In many Analog Devices application circuits, they show a 0.1 uF ceramic and 10 uF electrolytic in parallel right at the device pins.
Update-2:
Besides acting as an energy store for signal transient currents, C40 is the shunt leg of a lowpass filter. The series leg is the combination of resistances and inductances in the wires and pc board traces between the power source and C40, plus the output impedance of the power source itself. Note that these are complex impedances, not simple resistances; their values vary with frequency. You can expect them to be higher at a signal current frequency of 10 kHz compared to a signal current of 1 kHz.
These are very small impedance values, difficult to measure and messy to calculate; for normal circuits and wiring, less than 1 ohm. Because an electrolytic capacitor has a relatively high equivalent series inductance, its impedance increases at high frequencies. This is why good design practice is to add a second lowpass filter right at the IC power pin, this time with a ceramic capacitor that has a much higher self-resonant frequency. Some Analog Devices app notes (love me ADI docs!) show a 10 ohm resistor in series with the power source, to give the decoupling capacitor(s) a higher series resistance to work with.
A search for power supply decoupling (no quotation marks) returned many documents and videos in exactly 0.5 seconds. Like this:
https://www.eetimes.com/bypass-or-decouple-your-way-to-power-supply-noise-reduction/
And, of course, this:
https://www.analog.com/media/en/training-seminars/tutorials/MT-101.pdf
At the end are links to eleven (11) other documents on this topic.
