(the chip under test will) send a signal back to the
microcontroller hopefully at the same frequency (thats what I am
testing).
....
I will
analyze the frequency at certain points at calculate signal loss as a
percentage (assuming there is signal loss, since that is the point of
doing this test) compared to the number of ICs.

So, it woulds like you will be testing for frequency deviations or
'jitter' created by the ICs under test. If this is correct, and you
want to measure this linearly, then you can't use one microcontroller
and loop the signal in and out several times, because the outputs will
all be synchronised with the uC clock.

You also mention signal loss. If you are interested in amplitude, note
that the uC will always output 0 and 5V, no matter what level it gets on
the input. Maybe you mean modulated signal distortion?

So far it's not sounding to me like the AT megas are what you need to
do this project. Try using a simple function generator and a spectrum
analyzer.