I don't think it will split. BU only serves to not activate Segwit. Miners don't want BU either.
I know you have this miners are happy with the 1MB limit mantra, but does past actions on behalf of the miners support this?
Miners used to limit blocks to 250kB, then 500kB, then 750kB, and then 1MB. That is, in the past they have implemented a soft limit in policy.h or using -blockmaxsize command line option. If they wanted to squeeze fees higher, they could reduce the block size now. They can't increase it as they have hit the protocol consensus limit.
When these limits were increased, there wasn't yet a fee market. It was still a totally technical parameter that could be changed without too much hassle. In fact, as long as there is no pressure on the fee market, miners have tendency to INCREASE the block size in order to get more transactions, and hence more fees, fees that are very small.
After the last block reward halving, which coincided more or less with the emergence of a fee market (that is, sometimes full blocks and "fighting for a place") however, this block size became an economic parameter.
You are right that they could reduce block size now with a soft fork, diminishing the accepted block size, but *in order for a soft fork to succeed, it must have 51% of mining power*.
A miner voluntarily reducing HIS blocks to 512 KB he would just reap in less fees, augmenting the pressure on the market and increase the fees HIS competitors will reap in at double dose (1 MB).
If he only builds on blocks of 512 KB, and there's no 51% miner hash rate doing the same, all his blocks will be orphaned.
So:
1) there's no individual advantage to produce 512KB blocks, and forego half of the fees that 1MB blocks bring to competitors
2) trying to orphan 1MB blocks if you don't have more than 51% of hash power doing the same will only lead you, as a miner, to produce orphaned blocks.
3) the "common good" for miners, to reduce blocks even more, is offset by the tragedy of the commons. They first have to agree amongst themselves to apply this block reduction with more than 51% of hash rate, for this soft fork to succeed.
==> it won't happen.