That is a HUGE architecture change. In some installations that have common attachments, that would be a benefit. In other installations that would be a detriment because of the time it would take for lookups and orphans etc.
When looking towards the future and full encryption, the attachment is part of the message and couldn't be separated.
If you have a situation where your concerned about duplicate data then I would suggest you look at centralized storage solution that has de-duplication at the hardware level. This is really the only way to get the performance necessary to handle high disk i/o loads!
SmarterMail is about as efficient as possible. We built SmarterMail because we owned one of the largest Windows Hosting companies in the industry. We were hosting millions of users mailboxes 15 years ago and we couldn't find a stable mail server or a mail server that would fully utilize the hardware we were purchasing.
Every decision we have made is a balancing act to get the most out of server hardware and handle hundreds of thousands of transactions per second.
We even think about NAS solutions in relation to SSD Caches. For example, the way we organize our data in GRP files for a day allows active data to be on more expensive disks (SSD) where old data gets moved to legacy drives. This way you can purchase cost effective centralized storage solutions with high volume disk i/o with large storage capacities.
I appreciate you thinking this way and throwing out suggestions. What might look like a simple change normally has 100's of counter arguments from a programming and implementation perspective. Its kind of ridiculous how full our whiteboards get discussing a single checkbox setting let alone a huge architecture change!
Hope this helps,