Can't find any official reference to this, but one way to send OSC over
TCP was to prefix each packet with a 32-bit int (in network byte order
of course) giving the lenght of the packet [1]. IIRC there was even some
RFC about sending any type of packets over TCP by doing exactly that.
In 1.1 this was replaced by SLIP encoding, destroying the nice 32-bit
alignment of all data elements that we had before. Apparently this
was selected only because the OSC authors happened to have some
hardware using SLIP encoding -- not a good idea IMHO.

From the 1.1 paper (https://cnmat.berkeley.edu/publications/features-and-future-open-sound-control-version-11-nime) :

"This new specification enormously expands the range of protocols and hardware transports that can be used to communicate OSC encoded packets including Firewire, Ethernet, and USB (using TCP/IP); RS232 and RS422 and Serial USB and Serial Bluetooth and Serial Zigbee"

So it looks like using SLIP is a compromise. Maybe the method you mentionned for TCP could be added, but the overhead of SLIP seems to add only 4 bytes (32 bits): https://en.wikipedia.org/wiki/Serial_Line_Internet_Protocol

(disclaimer: I'm not a network specialist)

Marc