Why can’t it be 8?, the fact that it’s a trit doesn’t put any constraint on the byte (tryte ? size). You could actually make it 5 or 6 trits (~9.5 bits) for similar information density. The Setun used 6 trit addressable units.
Might be advantageous to collect trits in powers of 3, like how bits are collected in powers of 2.
For example, if you want to address a trit in a tryte of 8 trits, you can use 2 trits for that, but you'll have a value left over. Seems a bit cleaner if trytes hold 9 trits.