Is there any security reason not to encrypt directly Payload.zip with RSA public key designed for exchange ?
Technically, one could use RSA as a block cipher (like AES), encrypting data in blocks of (key size - padding) length and using a mode of operation to securely apply the same algorithm to multiple chunks of data. However, this would be dreadfully slow. RSA is an extremely expensive operation, computationally, especially if you use keys expected to be secure for decades to come (which you usually must, because the public key is, of course, public; unless the data will be valueless in a very short time, you need to use keys of quite expensive length). Fortunately, there's no need to ever do so. Every widely-used system that relies on RSA encryption is a hybrid cryptosystem, using expensive asymmetric crypto (such as RSA) to securely exchange a strong, cheap, and usually ephemeral (single-use) symmetric key (such as AES). Thus, the expensive RSA operation need only be performed one (at each end), and the rest can be done using a symmetric algorithm that is not only much faster in software, but also often can be accelerated by modern hardware.
How to compare security of RSA encryption between a (really small) 32 bytes (the AES key) file and a larger Zip file ?
32 bytes (256 bits) might sound really short, but that much pure entropy (randomness) is enough to secure something the length of a ZIP file from now until the heat death of the universe, even if you turn all of the earth into CPUs and were somehow able to power them for that long. 2^256 - the number of possible values for such a key - is a staggeringly large number; you could possibly assign a unique value to every single atom in the observable universe using only 256 bits.
Almost any RSA key is actually going to be much, much weaker than that. Unlike AES keys, RSA keys are nowhere near 100% entropy; most numbers of a given length aren't valid keys. A valid RSA key needs to be the product of two very large prime numbers. Most large numbers have more than two prime factors (and the proportion gets worse the bigger the numbers are), so to get twice the entropy (double the number of possible valid keys), you need to make an RSA key significantly more than twice as long. To get as much entropy as a 256-bit AES key, you'd need something like 17000 bits of RSA key.