Putting a finger on the antenna of the nRF24l01 cannot improve the filtering of the 3.3v power rail so you need to look at the issues separately.
When using capacitors on a power rail bigger does not necessarily mean better. The supply resistance and the capacitor have a cut-off frequency .. the larger the capacitor the lower the frequency. High value capacitors will suppress low frequency ripply but not high frequency. That is the reason for putting a 0.1uF in parallel with a 100uF to sort out high frequencies and low frequencies at the same time.
I am not convinced that the problem you are seeing is caused by the power supply. When you transmit for A to B the B end transmits to acknowledge the packets so both ends can transmit ok. If the transmission from B to A fails it may be related to the length of transmission as the transmit time in this case is much longer than just sending an ack.
Putting a finger on the antenna will not only make you act as an aerial it will seriously de-tune the transmitter circuit. It could be that this is sufficient to move the frequency of the transmitter. You have not said if you have tried other nRF24l01 modules other than the PA option. If it is just one module I would say it was faulty and not operating on the correct frequency.