That’s a great question. We thought it was the right way to go for now. Scaling out to multiple canisters has advantages (more memory, can run concurrently) and disadvantages (clearly more complex to deal with multiple canisters). So we start with the simpler approach, and can always transition to a multi-canister approach in the future.
Note that the UTXO set doesn’t grow super quickly Blockchain.com | Charts - Unspent Transaction Outputs. In the last 3 years, it grew from 64M UTXOs to 85M, and we plan to grow the stable memory further, so I don’t think the single canister approach will be a problem any time soon, if ever.