Should the most significant byte of a cryptograhic key be nonzero?

Is it safer to give a non zero value to the most significant byte (MSB) of a cryptographic key to ensure its length (not 31 instead of 32 for an EC private key on a 256-bit curve for example)?

Or is it counterproductive because it decreases the entropy?

I ask the question for the MSB but it could be for several of the first bytes.

Note that if you view the key value as a number that you would only have to set the MSB bit to 1 to ensure that the key value is at least $2^{(n-1)}$ - assuming that the key uses unsigned values of course.