The minority of miners that do not support BIP91 (and by extension BIP141), could start rejecting blocks of those who are signaling BIP141, thereby creating a different chain. Is that a possibility both in theory and in practice?

Yes that is possible, but unlikely. Since BIP 91 has a majority of the hashrate, I think it is unlikely that there will be a persistent chain fork. Rather what is more likely to happen is that there are short multi-block forks that are then orphaned and reorged out as the majority hash rate chain becomes longer than those forks.

Definition chain split: A state of the network in which there are two or more chaintips which are actively being extended with new blocks.

Chain splits can occur e.g. when:

• Two miners get lucky at the same time and publish a block at the same height.
• A HF gets activated on the network, which with the help of wipeout protection could become a persistent chain split.
• Rules in the network are not being consistently enforced, e.g. when a softfork is being activated but not enforced by all mining power. This is generally not persistent if a majority enforces the soft fork.

Currently, there are about 85-90% of the hashrate signalling for BIP91. A chain split could occur, if an unupgraded miner creates a non-signalling block NSB. In that case, miners enforcing BIP91 would ignore NSB and create another block at the same height. Meanwhile, any miners that do not enforce BIP91 (which unfortunately could perhaps include miners that did signal readiness for BIP91 but are not enforcing it) could be building on NSB. Until the cumulative difficulty in the signalling chaintip is greater than on the chain building on top of NSB, the network would be in a state of chain split.

Hopefully, all of the miners currently signalling readiness for BIP91 are actually enforcing BIP91, and any NSB would not find any succeeding blocks before being overtaken.