The SMTP RFC 5321 (see https://www.ietf.org/rfc/rfc5321.txt) specifies a sending server must queue messages which cannot be directly delivered because of a temporary failure at the receiving end.
The reason is that if it is a temporary issue with the email infrastructure, emails will not be bounced immediately but are instead queued on the sending server(s) and automatically retried for delivery.
In case of downtime of the destination mail server, messages are only accepted for delivery by the filter cluster if the recipient is known to be valid.
Valid destination recipients are cached for up to 96 hours.
When a destination server cannot be reached for 4 days, all messages will be bounced after 4 days and no new email will be accepted/queued until the destination server is back online.
This 4 day period is to conform with the SMTP RFC.
The reason it's not longer than 4 days, is because it's important for the sender to be aware that delivery of their message has been failing for 4 days so they can try and contact the recipient in another way.
