My shelter has this problem as well, and we don't have any easy solutions. Often bonded pairs are flagged to go to foster care earlier than single cats because we know they're less likely to be adopted, and sometimes once they're in foster care we receive more detailed information about how they behave in a home together. Owners are often very sentimental and will anthropomorphize these cats and say they're bonded to make themselves feel better, but then when we see them attacking one another in a foster home or not spending time together at all, we know we can safely unbond them.
Having said that, we still struggle with staff feeling that cats are bonded just because they came from the same home, or because they were housed together for an extended period of time. It's a common source of friction, particularly as unnecessary bonding can really extend length of stay.
I like David's suggestion of letting them settle in together and then housing them separately for a time to see if being apart impacts their eating/toileting habits. Have said that, we often have bonded pairs where one is fat and the other is slim, so it wouldn't surprise me if the data were skewed (like if the fat cat was eating ALL of the food the first few days, it would make it seem that the shy cat was on hunger strike once they were separated when she hadn't actually eaten at all since arriving at the shelter).
I feel like the best solution is to send them to foster homes early and often for better data, and to perform probing surrender interviews when an owner/agent for owner insists that a pair (or trio) of cats are bonded. We also offer a two for one discount on bonded cats, we do NOT bond kittens under 6 months, and if we become overwhelmed with bonded pairs we'll feature them heavily on social media and advertisements.
Ultimately, having data to work with is essential. If your data shows you that bonded pairs have a significantly longer length of stay, that they're more likely to become sick, and that they cost the shelter more money on average, then you'll want to use that data to show staff/volunteers that it's better for the cats to be separated so they can go home faster.