Runstest and corrcoef : difference and problems

Hi. I have noticed that if i use rand() function to generate for example 2*10^6 numbers and then i evaluate the correlation coefficient through corrcoef() i always obtain good results (null hypothesis not rejected and small values of the non diagonal elements of corr matrix).

On the contrary if i use runstest sometimes i get null hypothesis rejected and consequently small values of pvalue.

Is it a good test choice using runstest to test unif. random numbers?