The key differentiator between the recommended locations (e.g. local storage, a SAN, or an iSCSI-based network) and a network share is quite simply, redundancy. All of the recommended approaches provide an option for redundant paths for I/O to take to persistent storage.
For instance, you can RAID local storage, providing redundancy if any disk fails. With either the SAN or iSCSI-based network storage approaches, these technologies use Multipath Input/Output (MPIO) drivers, providing redundancy to the storage.
A network drive, in contrast, does not employ or allow for any redundant I/O paths. If a Network Interface Controller (NIC) fails on either end, the share likely disappears. Even if you have multiple NICs, there will still be a brief outage as a different IP address will now host that share, so any data sent to the old/failed IP will timeout and disappear. Basically a network share wasn't designed with this level of redundancy, and a loss of data mid stream may corrupt your database (or worse just get lost without a trace). The whole point of a database is to reliably store data and a network share brings that whole reliable aspect into question.