Since you have specifically asked about entropy, I'll start by covering the objective aspects of your question.
The entropy is reduced, because you are taking away half the keyspace from each side.
To simplify this, if the entropy of a 10 character password with lowercase and uppercase characters is
log2(52^10) = ~ 57 bits of entropy
then you remove the uppercase characters from the first half, and the lowercase characters from the last half, your entropy is effectively
log2(26^10) = ~ 47 bits of entropy
Plus 1 bit since that can occur the other way round too gives 48 bits. The above is assuming that the passwords are generated using a cryptographically secure pseudo random number generator - i.e. it is completely random and does not hold any patterns generated by the human brain that will be unintentionally predictable in many ways.
The entropy of a password is based on the generation technique, so if this is known then you have reduced the entropy as per the above. Don't rely on the method remaining secret.
However, if you want to do this for ease of entry, simply add two more characters to your password:
log2(26^12) = 56
Plus the 1 bit and you then have 57 bits for equivalent security of your original method.
However, you also asked about how this affects the practical security of your password. As said, unless the attacker knows the generation method of your password, in this case X number of characters that can be entered without the shift key and X number of characters that require the shift key, any password guessing attacks are going to be generic. That is, they are not targeting you in particular, and thus they do not know anything about your specific generation method. This means they will be trying generic word lists and common character sequences instead of sequences that fit your generation pattern. Therefore the entropy of their attack space does not match the entropy that your password generation method holds. This effectively makes your password practically more secure than the keyspace it is generated under in the case of non-targeted attacks.
In your case, assuming a randomly generated 20 character password this gives ~131 bits normally, and ~113 bits splitting it as described. Both are above the 80 bits NIST recommend for password strength.