The difference is negligible. Why does the 'T' parameter (symbol period) seem to have no effect at all? For these two, it's literally equal:

isequal(rcosfir(.2, 3, 12, 1/2400, 'sqrt'), rcosfir(.2, 3, 12, 1/9600, 'sqrt'))

#Matlab

Thanks. 

Notice that rcosfir generaly computes impulse response:

h(t) = SINC(t/T) COS(pi R t/T)/(1 - 4 R^2 t^2 /T^2)

at time instants t = n*T (where n is integer), which gives us

h(n) = SINC(n) COS(pi R n)/(1 - 4 R^2 n^2)

This, us you noticed, is independent from input signal sampling period. 

Thank you! Can you explain why they made it an input argument? It seems useless.

Not sure. Probably this resulted from continous time domain based approach to the design. Notice that currenlty, rcosdesign is recomended instead of rcosfir. The new script does not relate to continous time and sampling rate.

Just so you are aware, the raised cosine filter is not always the best choice for Nyquist filtering.

http://engineering.sdsu.edu/seshagir/SDR05.pdf